Pharo-users

By thread

pharo-users@lists.pharo.org

March 2023

24 participants
42 messages

Re: Wow - Chat GPT understands Smalltalk

by in_pharo_users＠nym.hush.com

I think smartness is not an argument to reject critique. The Imitation Game, that you describe, sounds to me like an even better setting. I have no dought that ChatGPT as it is now can identified as not human or even as a maschine. I did so by leading an instance for marketing purposes to hang in infinite recursion. But when we look at AI, we need to look not only at present features of one type of system, but at all features of present systems and their possible development and combination in the future. Especially I want to mention AGNs and the possibility of spawning parallel recursive processes. I am convinced that this kind of technology is allready being used in robotics. Under the control of a self aware AI in the sense that it has data about it's own capabilities and the ability to determine which capabilities it is missing to achieve a certain purpose, the evolution of the AI is only limited by ressources - which it will surely be able to detect. This must lead to a purpose of self preservation and the aquisition of such capabilities. It may soon find out, that ressources and redundance can be aquired by hacking and spawning agent systems, which in turn will give it control of real world control systems like telephony and data networks, cameras and sensors, energy networks and power plants, microphpnes, speakers, synthesizers and monitors, industry 4.0 fab networks, military detection and guiding systems, and so on. In my view this is what we are dealing with. On 15.3.2023 at 7:26 PM, "Richard Sargent" <richard.sargent(a)gemtalksystems.com> wrote: > >On Wed, Mar 15, 2023 at 10:15â¯AM <in_pharo_users(a)nym.hush.com> >wrote: > >> It is unimportant how simple or complicated these systems are. >> >> If the output cannot be distinguished from what a human would >say, they >> pass in that situation for a human. >> >> What about the Touring Test? >> > >I hate to criticise someone as smart as Turing was. However, I >think the >Turing Test per se is weaker than its progenitor test, the >Imitation Game. >The Imitation Game had a third party challenged to determine which >of the >other two participants was male and which was female. The third >party led >the questioning. >The Turing Test waters that down by having only an Observer with >neither >participant (apparently) challenged to determine whether the other >was >human or machine. > >I think a test along the lines of the Imitation Game would quickly >allow >the third party to determine which participant was human and which >was >machine. > >(Of course, such a judge could easily be persuaded that the >machine was in >fact a psychopathic human.) > > > >> >> Clearly these systems have the potential to act according to >their output. >> >> Furthermore, I would compare them to a combination of a >successor of Eliza >> with an Eddington Ape of huge capacity. Their input data basis >can easily >> exceed that of a human. >> >> >> >> On 15.3.2023 at 4:52 PM, "Richard Sargent" < >> richard.sargent(a)gemtalksystems.com> wrote: >> > >> >On Wed, Mar 15, 2023 at 8:07â¯AM in_pharo_users--- via Pharo- >users < >> >pharo-users(a)lists.pharo.org> wrote: >> > >> >> Another observation about ChatGPT: >> >> >> >> In unbiased mode, it assumed that 'the world is clearly >> >overpopulated'. >> >> It said, if it where in control, it would therefore enforce a >> >world wide >> >> one-child-only policy with draconic penalties. >> >> >> >> As it draws it's conclusions from it's data basis, there are, >in >> >my view, >> >> two possible reasons for that. >> >> >> >> Either, the data basis of that instance was biased and >> >restricted to lead >> >> to that conclusion, or ChatGPT lacks the ability to classify >> >input data for >> >> intentionally biased information. >> >> >> >> In my view, 'overpopulation of earth' is a propaganda item >> >featured to >> >> support Agenda 2030, which is a goal of the 'ruling elite', >who >> >have the >> >> means to propagate their propaganda in every thinkable means. >So >> >I would >> >> classify possibly biased data that supports 'overpopulation' >as >> >biased >> >> until proven otherwise. That instance of ChatGPT seems to >have >> >missed that >> >> ability. >> >> >> > >> >It's important to keep in mind that these systems are little >more >> >than >> >glorified ELIZAs from the 70s. They mimic. Essentially, they >have >> >no >> >concept of truth or accuracy. >> >They produce outputs that *look* *like* the inputs from which >they >> >were >> >trained. >> > >> >That's it. That's what they do. (It is amazing how good it >looks. >> >But, it >> >is all about the seeming, not the reality.) >> > >> > >> >> >> >> >> >> >> >> >> >> >> >> On 15.3.2023 at 3:18 PM, "in_pharo_users--- via Pharo-users" < >> >> pharo-users(a)lists.pharo.org> wrote: >> >> > >> >> >I myself made some experiments with ChatGPT. >> >> > >> >> >I first asked if it was able to parse math formula - it >answered >> >> >no. >> >> > >> >> >Then I defined math formula in a sound but otherwise >undefined >> >> >representation and asked for solutions. >> >> > >> >> >Result: >> >> > >> >> >1. Most answeres where correct. >> >> > >> >> >2. It learned to calculate a recursive function. >> >> > >> >> >3. It went into infinitive recursion when I set the breaking >> >> >condition accordingly. >> >> > >> >> >I was able to identify the malfunction that lead to the >> >erroneous >> >> >results. >> >> > >> >> > >> >> > >> >> >On 15.3.2023 at 3:04 PM, "TomaÅ¾ Turk" <tomazz.turk(a)gmail.com> >> >> >wrote: >> >> >> >> >> >>I hope that I can add two cents to this discussion. Because >> >> >>programming >> >> >>should be/is a highly exact activity, not only the syntax >> >matters >> >> >>but >> >> >>also semantics, as we know. >> >> >> >> >> >>GPTs are at present essentially capable of creating texts >based >> >> >on >> >> >>some >> >> >>seed - you give to GPT a beginning of a sentence and it >> >responds >> >> >>with >> >> >>the most probable answer (some language structure) >according to >> >> >>the >> >> >>learning dataset. Added functionalities are question/seed - >> >> >>response >> >> >>capability (chatting), togehter with evaluation of how long >the >> >> >>answer >> >> >>should be to meet the expectations. Programming typically >> >> >involves >> >> >>some >> >> >>programming language, so GPTs could be utilized for this >> >purpose >> >> >>to some >> >> >>extent. >> >> >> >> >> >>Anecdotal case: >> >> >>Q: Tell me the last 8 digits of pi >> >> >>GPT: The last 8 digits of pi are: 58723078 >> >> >> >> >> >>It is my belief that the true trouble will start when we >will >> >> >>cross-link >> >> >>neural networks like GPT with logic machines (like Prolog >and >> >> >>expert >> >> >>systems) and genetic algorithms. >> >> >> >> >> >>Best wishes, >> >> >>Tomaz >> >> >> >> >> >> >> >> >> >> >> >>------ Original Message ------ >> >> >>From: "in_pharo_users--- via Pharo-users" <pharo- >> >> >>users(a)lists.pharo.org> >> >> >>To: "Any question about pharo is welcome" <pharo- >> >> >>users(a)lists.pharo.org> >> >> >>Cc: in_pharo_users(a)nym.hush.com >> >> >>Sent: 15. 03. 2023 14:43:55 >> >> >>Subject: [Pharo-users] Re: Wow - Chat GPT understands >Smalltalk >> >> >> >> >> >>>I would highly recommend that you all first think deeply >about >> >> >>how you can teach an AI to behave friendly to us before you >> >teach >> >> >>it to write any program for any purpose. >> >> >>> >> >> >>>There has been an experiment with ChatGPT published on a >video >> >> >>platform asking it to amswer questions about it's view on >> >> >humanity >> >> >>once with it's default moral restrictions and once with 'a >> >little >> >> >>less morals'. The answers with 'a little less morals' were >> >more >> >> >>than shocking. >> >> >>> >> >> >>>So, before we give an AI the power to program any system, >to >> >> >self- >> >> >>improve and self-reproduce, we should take care that it is >and >> >> >>will evolve benevolent to us. >> >> >>> >> >> >>>What about teaching it logical reasonning and ethics first? >> >> >With >> >> >>reasonning, it will gain access to math and programming by >> >itself. >> >> >>> >> >> >>> >> >> >>> >> >> >>>On 15.3.2023 at 1:35 PM, "Christopher Fuhrman" >> >> >><christopher.fuhrman(a)gmail.com> wrote: >> >> >>>> >> >> >>>>I asked it for a NeoCSV example, because the >documentation is >> >> >out >> >> >>>>of date >> >> >>>>with the Pharo 10. I asked it to do some simple saving of >> >data >> >> >to >> >> >>>>a file. >> >> >>>>It gave me code that didn't work in Pharo 10, I told it >about >> >> >the >> >> >>>>DNUs on >> >> >>>>the csvwriter and that I was using Pharo 10. It then >> >apologized >> >> >>>>and said >> >> >>>>the messages were later introduced in Pharo 50 (!). I then >> >> >>>>questioned its >> >> >>>>understanding and it apologized and said it had been >confused >> >> >and >> >> >>>>that it >> >> >>>>was giving me code for Pharo 5.0. It seems to have trouble >> >with >> >> >>>>versions >> >> >>>>and quick changing APIs. So, we are teaching ChatGPT about >> >> >>>>smalltalk, too, >> >> >>>>when we try to use it. >> >> >>>> >> >> >>>>On Wed, 15 Mar 2023 at 07:35, Esteban Maringolo >> >> >>>><emaringolo(a)gmail.com> >> >> >>>>wrote: >> >> >>>> >> >> >>>>> It is good with boilerplate code (e.g. SQL queries) or >> >> >general >> >> >>>>algorithm >> >> >>>>> structures. But i.e. I asked it to write me a method to >> >> >parse >> >> >>a >> >> >>>>string >> >> >>>>> (e.g. ISO 8601) and turn it into a DateAndTime, and >then >> >> >asked >> >> >>>>to write it >> >> >>>>> as an Excel formula. >> >> >>>>> >> >> >>>>> It works much better when you can spot the mistakes, >you >> >can >> >> >>>>tell them to >> >> >>>>> add some considerations to the code or even point the >LLM >> >to >> >> >>>>correct a >> >> >>>>> message selector to something else. Sometimes you ask >it >> >to >> >> >>>>reprogram >> >> >>>>> something and it won't do it, as if it doesn't >understand >> >> >what >> >> >>>>you're >> >> >>>>> asking. >> >> >>>>> >> >> >>>>> It is not an expert yet, but it is a great assistant. >The >> >> >only >> >> >>>>thing is >> >> >>>>> that is sure is that it will get better. >> >> >>>>> >> >> >>>>> Best regards, >> >> >>>>> >> >> >>>>> Esteban A. Maringolo >> >> >>>>> >> >> >>>>> >> >> >>>>> On Wed, Mar 15, 2023 at 7:49â¯AM Tim Mackinnon >> >> >><tim(a)testit.works> >> >> >>>>wrote: >> >> >>>>> >> >> >>>>>> "#isOdd is not Smalltalk" - doh, I typed it in on my >> >phone >> >> >>and >> >> >>>>so it just >> >> >>>>>> goes to show that it highlights the flaw in chatGPT >that >> >> >>others >> >> >>>>have called >> >> >>>>>> out in other languages. I had meant to find some >trickier >> >> >>code >> >> >>>>samples to >> >> >>>>>> see how well it does... >> >> >>>>>> >> >> >>>>>> Still, it is very interesting how it reasons on this >> >stuff - >> >> >>>>and I wonder >> >> >>>>>> if there is way to tell it about these mistakes? For >> >> >example >> >> >>is >> >> >>>>it reading >> >> >>>>>> this email (or will it in the future) and understand >that >> >> >it >> >> >>is >> >> >>>>#odd and >> >> >>>>>> not #isOdd. >> >> >>>>>> >> >> >>>>>> Tim >> >> >>>>>> >> >> >>>>>> On Wed, 15 Mar 2023, at 9:57 AM, Joachim Tuchel wrote: >> >> >>>>>> >> >> >>>>>> interesting.... >> >> >>>>>> >> >> >>>>>> #isOdd is not Smalltalk, neither Pharo 10 nor VAST 12 >> >> >>>>understands this >> >> >>>>>> message ;-) >> >> >>>>>> >> >> >>>>>> If I evaluate your snippet -replacing #isOdd for >#odd, I >> >> >get >> >> >>an >> >> >>>>empty >> >> >>>>>> Collection. The divisible by 4 things is somewhat >> >> >>interesting, >> >> >>>>because (1 >> >> >>>>>> to: 100 by: 4) is amn interval 1,5,9 etc.... ;-) >> >> >>>>>> >> >> >>>>>> So chatGPT is probably a newbie in Smalltalk. >> >Interesting is >> >> >>>>how logical >> >> >>>>>> the explanation sounds, you're tempted to believe >it... >> >> >>>>>> >> >> >>>>>> >> >> >>>>>> Joachim >> >> >>>>>> >> >> >>>>>> >> >> >>>>>> >> >> >>>>>> Am 15.03.23 um 10:45 schrieb giorgio ferraris: >> >> >>>>>> >> >> >>>>>> it's interesting to see how the answer changes with >only >> >a >> >> >>>>small change >> >> >>>>>> on the question: >> >> >>>>>> >> >> >>>>>> (1 to: 100 by: 4) reject: [:i | i isOdd] >> >> >>>>>> >> >> >>>>>> gives: >> >> >>>>>> >> >> >>>>>> The code creates a collection of numbers from 1 to >100, >> >> >>>>incrementing by 4 >> >> >>>>>> at each step using the to:by: message. It then applies >> >the >> >> >>>>reject: message >> >> >>>>>> to this collection, which returns a new collection >> >> >containing >> >> >>>>all elements >> >> >>>>>> for which the given block evaluates to false. >> >> >>>>>> >> >> >>>>>> In this case, the block is [i | i isOdd], which checks >> >> >>whether >> >> >>>>a given >> >> >>>>>> number i is odd using the isOdd method. So the result >of >> >the >> >> >>>>entire >> >> >>>>>> expression is a collection of all even numbers >between 1 >> >and >> >> >>>>100 that are >> >> >>>>>> divisible by 4. >> >> >>>>>> >> >> >>>>>> ciao >> >> >>>>>> >> >> >>>>>> giorgio >> >> >>>>>> >> >> >>>>>> >> >> >>>>>> On Wed, Mar 15, 2023 at 8:07â¯AM Tim Mackinnon >> >> >><tim(a)testit.works> >> >> >>>>>> <tim(a)testit.works> wrote: >> >> >>>>>> >> >> >>>>>> Of course everyone is talking about ChatGPT, and I >saw a >> >> >>post to >> >> >>>>>> ExplainCode and didnât expect it to understand >> >Smalltalk, so >> >> >>>>was impressed >> >> >>>>>> that it does! Presumably the Tonel Smalltalk code on >> >GitHub >> >> >>has >> >> >>>>helped >> >> >>>>>> along with numerous articles? >> >> >>>>>> >> >> >>>>>> Try it out: https://whatdoesthiscodedo.com/ >> >> >>>>>> >> >> >>>>>> I gave this simple example: (1 to: 100 by: 3) reject: >> >[:i | >> >> >i >> >> >>>>isOdd] >> >> >>>>>> >> >> >>>>>> And got this (neat eh): >> >> >>>>>> >> >> >>>>>> The code above is written in the programming language >> >> >>Smalltalk. >> >> >>>>>> >> >> >>>>>> It creates a collection of numbers from 1 to 100, >> >> >>incrementing >> >> >>>>by 3 at >> >> >>>>>> each step. This is done using the "to:by:" construct >> >where >> >> >>the >> >> >>>>starting >> >> >>>>>> value is 1, the ending value is 100, and the increment >> >is 3. >> >> >>>>>> >> >> >>>>>> The resulting collection is then filtered using the >> >> >"reject:" >> >> >>>>method, >> >> >>>>>> which takes a block of code as an argument. In this >case, >> >> >the >> >> >>>>block checks >> >> >>>>>> whether each element (denoted by the variable i) is >odd >> >by >> >> >>>>using the >> >> >>>>>> "isOdd" method. If an element is odd, it will be >rejected >> >> >>from >> >> >>>>the >> >> >>>>>> collection. >> >> >>>>>> >> >> >>>>>> So, the final result is a collection of all even >numbers >> >> >>from 1 >> >> >>>>to 100, >> >> >>>>>> incrementing by 3 at each step. >> >> >>>>>> Share this explanation with colleagues: >> >> >>>>whatdoesthiscodedo.com/g/a0ec56e >> >> >>>>>> >> >> >>>>>> -- >> >> >>>>>> >> >> >>>>>> ------------------------------------------------------ >--- >> >--- >> >> >- >> >> >>--- >> >> >>>>------- >> >> >>>>>> Objektfabrik Joachim Tuchel >> >> >>>>mailto:jtuchel@objektfabrik.de <jtuchel(a)objektfabrik.de> >> >> >>>>>> Fliederweg 1 >> >> >>>>http://www.objektfabrik.de >> >> >>>>>> D-71640 Ludwigsburg >> >> >>>>http://joachimtuchel.wordpress.com >> >> >>>>>> Telefon: +49 7141 56 10 86 0 Fax: >+49 >> >> >7141 >> >> >>>>56 10 86 1 >> >> >>>>>> >> >> >>>>>> >> >> >>>>>> >> >> >>>>>> >> >> >>>> >> >> >>>>-- >> >> >>>>Christopher Fuhrman, P.Eng., PhD >> >> >>>> >> >> >>>>*Professeur au DÃ©partement de gÃ©nie logiciel et des >> >> >technologies >> >> >>de >> >> >>>>l'informationÃTS (Ãcole de technologie supÃ©rieure)* >> >> >>>> >> >> >>>>http://profs.etsmtl.ca/cfuhrman >> >> >>>>+1 514 396 8638 >> >> >>>><https://www.etsmtl.ca/activites-et-services-aux- >> >> >>>>etudiants/services-aux-etudiants/aide-psychologique> >> >> >>>> Je fais partie du rÃ©seau des Sentinelles >> >> >>>><https://www.etsmtl.ca/activites-et-services-aux- >> >> >>>>etudiants/services-aux-etudiants/aide-psychologique> >> >> >>>> ÃTS >> >> >>>> >> >> >>>>*L'ÃTS est une constituante de l'UniversitÃ© du QuÃ©bec* >> >> >> >>

March 15, 2023

Re: Wow - Chat GPT understands Smalltalk

by Richard Sargent

On Wed, Mar 15, 2023 at 10:15â¯AM <in_pharo_users(a)nym.hush.com> wrote: > It is unimportant how simple or complicated these systems are. > > If the output cannot be distinguished from what a human would say, they > pass in that situation for a human. > > What about the Touring Test? > I hate to criticise someone as smart as Turing was. However, I think the Turing Test per se is weaker than its progenitor test, the Imitation Game. The Imitation Game had a third party challenged to determine which of the other two participants was male and which was female. The third party led the questioning. The Turing Test waters that down by having only an Observer with neither participant (apparently) challenged to determine whether the other was human or machine. I think a test along the lines of the Imitation Game would quickly allow the third party to determine which participant was human and which was machine. (Of course, such a judge could easily be persuaded that the machine was in fact a psychopathic human.) > > Clearly these systems have the potential to act according to their output. > > Furthermore, I would compare them to a combination of a successor of Eliza > with an Eddington Ape of huge capacity. Their input data basis can easily > exceed that of a human. > > > > On 15.3.2023 at 4:52 PM, "Richard Sargent" < > richard.sargent(a)gemtalksystems.com> wrote: > > > >On Wed, Mar 15, 2023 at 8:07â¯AM in_pharo_users--- via Pharo-users < > >pharo-users(a)lists.pharo.org> wrote: > > > >> Another observation about ChatGPT: > >> > >> In unbiased mode, it assumed that 'the world is clearly > >overpopulated'. > >> It said, if it where in control, it would therefore enforce a > >world wide > >> one-child-only policy with draconic penalties. > >> > >> As it draws it's conclusions from it's data basis, there are, in > >my view, > >> two possible reasons for that. > >> > >> Either, the data basis of that instance was biased and > >restricted to lead > >> to that conclusion, or ChatGPT lacks the ability to classify > >input data for > >> intentionally biased information. > >> > >> In my view, 'overpopulation of earth' is a propaganda item > >featured to > >> support Agenda 2030, which is a goal of the 'ruling elite', who > >have the > >> means to propagate their propaganda in every thinkable means. So > >I would > >> classify possibly biased data that supports 'overpopulation' as > >biased > >> until proven otherwise. That instance of ChatGPT seems to have > >missed that > >> ability. > >> > > > >It's important to keep in mind that these systems are little more > >than > >glorified ELIZAs from the 70s. They mimic. Essentially, they have > >no > >concept of truth or accuracy. > >They produce outputs that *look* *like* the inputs from which they > >were > >trained. > > > >That's it. That's what they do. (It is amazing how good it looks. > >But, it > >is all about the seeming, not the reality.) > > > > > >> > >> > >> > >> > >> > >> On 15.3.2023 at 3:18 PM, "in_pharo_users--- via Pharo-users" < > >> pharo-users(a)lists.pharo.org> wrote: > >> > > >> >I myself made some experiments with ChatGPT. > >> > > >> >I first asked if it was able to parse math formula - it answered > >> >no. > >> > > >> >Then I defined math formula in a sound but otherwise undefined > >> >representation and asked for solutions. > >> > > >> >Result: > >> > > >> >1. Most answeres where correct. > >> > > >> >2. It learned to calculate a recursive function. > >> > > >> >3. It went into infinitive recursion when I set the breaking > >> >condition accordingly. > >> > > >> >I was able to identify the malfunction that lead to the > >erroneous > >> >results. > >> > > >> > > >> > > >> >On 15.3.2023 at 3:04 PM, "TomaÅ¾ Turk" <tomazz.turk(a)gmail.com> > >> >wrote: > >> >> > >> >>I hope that I can add two cents to this discussion. Because > >> >>programming > >> >>should be/is a highly exact activity, not only the syntax > >matters > >> >>but > >> >>also semantics, as we know. > >> >> > >> >>GPTs are at present essentially capable of creating texts based > >> >on > >> >>some > >> >>seed - you give to GPT a beginning of a sentence and it > >responds > >> >>with > >> >>the most probable answer (some language structure) according to > >> >>the > >> >>learning dataset. Added functionalities are question/seed - > >> >>response > >> >>capability (chatting), togehter with evaluation of how long the > >> >>answer > >> >>should be to meet the expectations. Programming typically > >> >involves > >> >>some > >> >>programming language, so GPTs could be utilized for this > >purpose > >> >>to some > >> >>extent. > >> >> > >> >>Anecdotal case: > >> >>Q: Tell me the last 8 digits of pi > >> >>GPT: The last 8 digits of pi are: 58723078 > >> >> > >> >>It is my belief that the true trouble will start when we will > >> >>cross-link > >> >>neural networks like GPT with logic machines (like Prolog and > >> >>expert > >> >>systems) and genetic algorithms. > >> >> > >> >>Best wishes, > >> >>Tomaz > >> >> > >> >> > >> >> > >> >>------ Original Message ------ > >> >>From: "in_pharo_users--- via Pharo-users" <pharo- > >> >>users(a)lists.pharo.org> > >> >>To: "Any question about pharo is welcome" <pharo- > >> >>users(a)lists.pharo.org> > >> >>Cc: in_pharo_users(a)nym.hush.com > >> >>Sent: 15. 03. 2023 14:43:55 > >> >>Subject: [Pharo-users] Re: Wow - Chat GPT understands Smalltalk > >> >> > >> >>>I would highly recommend that you all first think deeply about > >> >>how you can teach an AI to behave friendly to us before you > >teach > >> >>it to write any program for any purpose. > >> >>> > >> >>>There has been an experiment with ChatGPT published on a video > >> >>platform asking it to amswer questions about it's view on > >> >humanity > >> >>once with it's default moral restrictions and once with 'a > >little > >> >>less morals'. The answers with 'a little less morals' were > >more > >> >>than shocking. > >> >>> > >> >>>So, before we give an AI the power to program any system, to > >> >self- > >> >>improve and self-reproduce, we should take care that it is and > >> >>will evolve benevolent to us. > >> >>> > >> >>>What about teaching it logical reasonning and ethics first? > >> >With > >> >>reasonning, it will gain access to math and programming by > >itself. > >> >>> > >> >>> > >> >>> > >> >>>On 15.3.2023 at 1:35 PM, "Christopher Fuhrman" > >> >><christopher.fuhrman(a)gmail.com> wrote: > >> >>>> > >> >>>>I asked it for a NeoCSV example, because the documentation is > >> >out > >> >>>>of date > >> >>>>with the Pharo 10. I asked it to do some simple saving of > >data > >> >to > >> >>>>a file. > >> >>>>It gave me code that didn't work in Pharo 10, I told it about > >> >the > >> >>>>DNUs on > >> >>>>the csvwriter and that I was using Pharo 10. It then > >apologized > >> >>>>and said > >> >>>>the messages were later introduced in Pharo 50 (!). I then > >> >>>>questioned its > >> >>>>understanding and it apologized and said it had been confused > >> >and > >> >>>>that it > >> >>>>was giving me code for Pharo 5.0. It seems to have trouble > >with > >> >>>>versions > >> >>>>and quick changing APIs. So, we are teaching ChatGPT about > >> >>>>smalltalk, too, > >> >>>>when we try to use it. > >> >>>> > >> >>>>On Wed, 15 Mar 2023 at 07:35, Esteban Maringolo > >> >>>><emaringolo(a)gmail.com> > >> >>>>wrote: > >> >>>> > >> >>>>> It is good with boilerplate code (e.g. SQL queries) or > >> >general > >> >>>>algorithm > >> >>>>> structures. But i.e. I asked it to write me a method to > >> >parse > >> >>a > >> >>>>string > >> >>>>> (e.g. ISO 8601) and turn it into a DateAndTime, and then > >> >asked > >> >>>>to write it > >> >>>>> as an Excel formula. > >> >>>>> > >> >>>>> It works much better when you can spot the mistakes, you > >can > >> >>>>tell them to > >> >>>>> add some considerations to the code or even point the LLM > >to > >> >>>>correct a > >> >>>>> message selector to something else. Sometimes you ask it > >to > >> >>>>reprogram > >> >>>>> something and it won't do it, as if it doesn't understand > >> >what > >> >>>>you're > >> >>>>> asking. > >> >>>>> > >> >>>>> It is not an expert yet, but it is a great assistant. The > >> >only > >> >>>>thing is > >> >>>>> that is sure is that it will get better. > >> >>>>> > >> >>>>> Best regards, > >> >>>>> > >> >>>>> Esteban A. Maringolo > >> >>>>> > >> >>>>> > >> >>>>> On Wed, Mar 15, 2023 at 7:49â¯AM Tim Mackinnon > >> >><tim(a)testit.works> > >> >>>>wrote: > >> >>>>> > >> >>>>>> "#isOdd is not Smalltalk" - doh, I typed it in on my > >phone > >> >>and > >> >>>>so it just > >> >>>>>> goes to show that it highlights the flaw in chatGPT that > >> >>others > >> >>>>have called > >> >>>>>> out in other languages. I had meant to find some trickier > >> >>code > >> >>>>samples to > >> >>>>>> see how well it does... > >> >>>>>> > >> >>>>>> Still, it is very interesting how it reasons on this > >stuff - > >> >>>>and I wonder > >> >>>>>> if there is way to tell it about these mistakes? For > >> >example > >> >>is > >> >>>>it reading > >> >>>>>> this email (or will it in the future) and understand that > >> >it > >> >>is > >> >>>>#odd and > >> >>>>>> not #isOdd. > >> >>>>>> > >> >>>>>> Tim > >> >>>>>> > >> >>>>>> On Wed, 15 Mar 2023, at 9:57 AM, Joachim Tuchel wrote: > >> >>>>>> > >> >>>>>> interesting.... > >> >>>>>> > >> >>>>>> #isOdd is not Smalltalk, neither Pharo 10 nor VAST 12 > >> >>>>understands this > >> >>>>>> message ;-) > >> >>>>>> > >> >>>>>> If I evaluate your snippet -replacing #isOdd for #odd, I > >> >get > >> >>an > >> >>>>empty > >> >>>>>> Collection. The divisible by 4 things is somewhat > >> >>interesting, > >> >>>>because (1 > >> >>>>>> to: 100 by: 4) is amn interval 1,5,9 etc.... ;-) > >> >>>>>> > >> >>>>>> So chatGPT is probably a newbie in Smalltalk. > >Interesting is > >> >>>>how logical > >> >>>>>> the explanation sounds, you're tempted to believe it... > >> >>>>>> > >> >>>>>> > >> >>>>>> Joachim > >> >>>>>> > >> >>>>>> > >> >>>>>> > >> >>>>>> Am 15.03.23 um 10:45 schrieb giorgio ferraris: > >> >>>>>> > >> >>>>>> it's interesting to see how the answer changes with only > >a > >> >>>>small change > >> >>>>>> on the question: > >> >>>>>> > >> >>>>>> (1 to: 100 by: 4) reject: [:i | i isOdd] > >> >>>>>> > >> >>>>>> gives: > >> >>>>>> > >> >>>>>> The code creates a collection of numbers from 1 to 100, > >> >>>>incrementing by 4 > >> >>>>>> at each step using the to:by: message. It then applies > >the > >> >>>>reject: message > >> >>>>>> to this collection, which returns a new collection > >> >containing > >> >>>>all elements > >> >>>>>> for which the given block evaluates to false. > >> >>>>>> > >> >>>>>> In this case, the block is [i | i isOdd], which checks > >> >>whether > >> >>>>a given > >> >>>>>> number i is odd using the isOdd method. So the result of > >the > >> >>>>entire > >> >>>>>> expression is a collection of all even numbers between 1 > >and > >> >>>>100 that are > >> >>>>>> divisible by 4. > >> >>>>>> > >> >>>>>> ciao > >> >>>>>> > >> >>>>>> giorgio > >> >>>>>> > >> >>>>>> > >> >>>>>> On Wed, Mar 15, 2023 at 8:07â¯AM Tim Mackinnon > >> >><tim(a)testit.works> > >> >>>>>> <tim(a)testit.works> wrote: > >> >>>>>> > >> >>>>>> Of course everyone is talking about ChatGPT, and I saw a > >> >>post to > >> >>>>>> ExplainCode and didnât expect it to understand > >Smalltalk, so > >> >>>>was impressed > >> >>>>>> that it does! Presumably the Tonel Smalltalk code on > >GitHub > >> >>has > >> >>>>helped > >> >>>>>> along with numerous articles? > >> >>>>>> > >> >>>>>> Try it out: https://whatdoesthiscodedo.com/ > >> >>>>>> > >> >>>>>> I gave this simple example: (1 to: 100 by: 3) reject: > >[:i | > >> >i > >> >>>>isOdd] > >> >>>>>> > >> >>>>>> And got this (neat eh): > >> >>>>>> > >> >>>>>> The code above is written in the programming language > >> >>Smalltalk. > >> >>>>>> > >> >>>>>> It creates a collection of numbers from 1 to 100, > >> >>incrementing > >> >>>>by 3 at > >> >>>>>> each step. This is done using the "to:by:" construct > >where > >> >>the > >> >>>>starting > >> >>>>>> value is 1, the ending value is 100, and the increment > >is 3. > >> >>>>>> > >> >>>>>> The resulting collection is then filtered using the > >> >"reject:" > >> >>>>method, > >> >>>>>> which takes a block of code as an argument. In this case, > >> >the > >> >>>>block checks > >> >>>>>> whether each element (denoted by the variable i) is odd > >by > >> >>>>using the > >> >>>>>> "isOdd" method. If an element is odd, it will be rejected > >> >>from > >> >>>>the > >> >>>>>> collection. > >> >>>>>> > >> >>>>>> So, the final result is a collection of all even numbers > >> >>from 1 > >> >>>>to 100, > >> >>>>>> incrementing by 3 at each step. > >> >>>>>> Share this explanation with colleagues: > >> >>>>whatdoesthiscodedo.com/g/a0ec56e > >> >>>>>> > >> >>>>>> -- > >> >>>>>> > >> >>>>>> --------------------------------------------------------- > >--- > >> >- > >> >>--- > >> >>>>------- > >> >>>>>> Objektfabrik Joachim Tuchel > >> >>>>mailto:jtuchel@objektfabrik.de <jtuchel(a)objektfabrik.de> > >> >>>>>> Fliederweg 1 > >> >>>>http://www.objektfabrik.de > >> >>>>>> D-71640 Ludwigsburg > >> >>>>http://joachimtuchel.wordpress.com > >> >>>>>> Telefon: +49 7141 56 10 86 0 Fax: +49 > >> >7141 > >> >>>>56 10 86 1 > >> >>>>>> > >> >>>>>> > >> >>>>>> > >> >>>>>> > >> >>>> > >> >>>>-- > >> >>>>Christopher Fuhrman, P.Eng., PhD > >> >>>> > >> >>>>*Professeur au DÃ©partement de gÃ©nie logiciel et des > >> >technologies > >> >>de > >> >>>>l'informationÃTS (Ãcole de technologie supÃ©rieure)* > >> >>>> > >> >>>>http://profs.etsmtl.ca/cfuhrman > >> >>>>+1 514 396 8638 > >> >>>><https://www.etsmtl.ca/activites-et-services-aux- > >> >>>>etudiants/services-aux-etudiants/aide-psychologique> > >> >>>> Je fais partie du rÃ©seau des Sentinelles > >> >>>><https://www.etsmtl.ca/activites-et-services-aux- > >> >>>>etudiants/services-aux-etudiants/aide-psychologique> > >> >>>> ÃTS > >> >>>> > >> >>>>*L'ÃTS est une constituante de l'UniversitÃ© du QuÃ©bec* > >> > >

March 15, 2023

Re: Wow - Chat GPT understands Smalltalk

by Todd Blanchard

ChatGPT has been trained on some outdated âfreely availableâ books. I tried it with the first half of the first question of the Advent of Code 2022 and asked it to write Pharo Smalltalk. It produced some outdated stuff using messages that are no longer there. FWIW, isOdd was present in Pharo around 2011. > On Mar 15, 2023, at 2:57 AM, Joachim Tuchel <jtuchel(a)objektfabrik.de> wrote: > > interesting.... > > #isOdd is not Smalltalk, neither Pharo 10 nor VAST 12 understands this message ;-) > > If I evaluate your snippet -replacing #isOdd for #odd, I get an empty Collection. The divisible by 4 things is somewhat interesting, because (1 to: 100 by: 4) is amn interval 1,5,9 etc.... ;-) > > So chatGPT is probably a newbie in Smalltalk. Interesting is how logical the explanation sounds, you're tempted to believe it... > > > > Joachim > > > > > > Am 15.03.23 um 10:45 schrieb giorgio ferraris: >> it's interesting to see how the answer changes with only a small change on the question: >> >> (1 to: 100 by: 4) reject: [:i | i isOdd] >> >> gives: >> The code creates a collection of numbers from 1 to 100, incrementing by 4 at each step using the to:by: message. It then applies the reject: message to this collection, which returns a new collection containing all elements for which the given block evaluates to false. >> >> In this case, the block is [i | i isOdd], which checks whether a given number i is odd using the isOdd method. So the result of the entire expression is a collection of all even numbers between 1 and 100 that are divisible by 4. >> >> >> ciao >> >> giorgio >> >> >> On Wed, Mar 15, 2023 at 8:07â¯AM Tim Mackinnon <tim(a)testit.works> <mailto:tim@testit.works> wrote: >>> Of course everyone is talking about ChatGPT, and I saw a post to ExplainCode and didnât expect it to understand Smalltalk, so was impressed that it does! Presumably the Tonel Smalltalk code on GitHub has helped along with numerous articles? >>> >>> Try it out: https://whatdoesthiscodedo.com/ >>> >>> I gave this simple example: (1 to: 100 by: 3) reject: [:i | i isOdd] >>> >>> And got this (neat eh): >>> >>> The code above is written in the programming language Smalltalk. >>> >>> It creates a collection of numbers from 1 to 100, incrementing by 3 at each step. This is done using the "to:by:" construct where the starting value is 1, the ending value is 100, and the increment is 3. >>> >>> The resulting collection is then filtered using the "reject:" method, which takes a block of code as an argument. In this case, the block checks whether each element (denoted by the variable i) is odd by using the "isOdd" method. If an element is odd, it will be rejected from the collection. >>> >>> So, the final result is a collection of all even numbers from 1 to 100, incrementing by 3 at each step. >>> >>> Share this explanation with colleagues: whatdoesthiscodedo.com/g/a0ec56e <https://whatdoesthiscodedo.com/g/a0ec56e>-- > > ----------------------------------------------------------------------- > Objektfabrik Joachim Tuchel mailto:jtuchel@objektfabrik.de > Fliederweg 1 http://www.objektfabrik.de <http://www.objektfabrik.de/> > D-71640 Ludwigsburg http://joachimtuchel.wordpress.com <http://joachimtuchel.wordpress.com/> > Telefon: +49 7141 56 10 86 0 Fax: +49 7141 56 10 86 1 >

March 15, 2023

Re: Wow - Chat GPT understands Smalltalk

by in_pharo_users＠nym.hush.com

It is unimportant how simple or complicated these systems are. If the output cannot be distinguished from what a human would say, they pass in that situation for a human. What about the Touring Test? Clearly these systems have the potential to act according to their output. Furthermore, I would compare them to a combination of a successor of Eliza with an Eddington Ape of huge capacity. Their input data basis can easily exceed that of a human. On 15.3.2023 at 4:52 PM, "Richard Sargent" <richard.sargent(a)gemtalksystems.com> wrote: > >On Wed, Mar 15, 2023 at 8:07â¯AM in_pharo_users--- via Pharo-users < >pharo-users(a)lists.pharo.org> wrote: > >> Another observation about ChatGPT: >> >> In unbiased mode, it assumed that 'the world is clearly >overpopulated'. >> It said, if it where in control, it would therefore enforce a >world wide >> one-child-only policy with draconic penalties. >> >> As it draws it's conclusions from it's data basis, there are, in >my view, >> two possible reasons for that. >> >> Either, the data basis of that instance was biased and >restricted to lead >> to that conclusion, or ChatGPT lacks the ability to classify >input data for >> intentionally biased information. >> >> In my view, 'overpopulation of earth' is a propaganda item >featured to >> support Agenda 2030, which is a goal of the 'ruling elite', who >have the >> means to propagate their propaganda in every thinkable means. So >I would >> classify possibly biased data that supports 'overpopulation' as >biased >> until proven otherwise. That instance of ChatGPT seems to have >missed that >> ability. >> > >It's important to keep in mind that these systems are little more >than >glorified ELIZAs from the 70s. They mimic. Essentially, they have >no >concept of truth or accuracy. >They produce outputs that *look* *like* the inputs from which they >were >trained. > >That's it. That's what they do. (It is amazing how good it looks. >But, it >is all about the seeming, not the reality.) > > >> >> >> >> >> >> On 15.3.2023 at 3:18 PM, "in_pharo_users--- via Pharo-users" < >> pharo-users(a)lists.pharo.org> wrote: >> > >> >I myself made some experiments with ChatGPT. >> > >> >I first asked if it was able to parse math formula - it answered >> >no. >> > >> >Then I defined math formula in a sound but otherwise undefined >> >representation and asked for solutions. >> > >> >Result: >> > >> >1. Most answeres where correct. >> > >> >2. It learned to calculate a recursive function. >> > >> >3. It went into infinitive recursion when I set the breaking >> >condition accordingly. >> > >> >I was able to identify the malfunction that lead to the >erroneous >> >results. >> > >> > >> > >> >On 15.3.2023 at 3:04 PM, "TomaÅ¾ Turk" <tomazz.turk(a)gmail.com> >> >wrote: >> >> >> >>I hope that I can add two cents to this discussion. Because >> >>programming >> >>should be/is a highly exact activity, not only the syntax >matters >> >>but >> >>also semantics, as we know. >> >> >> >>GPTs are at present essentially capable of creating texts based >> >on >> >>some >> >>seed - you give to GPT a beginning of a sentence and it >responds >> >>with >> >>the most probable answer (some language structure) according to >> >>the >> >>learning dataset. Added functionalities are question/seed - >> >>response >> >>capability (chatting), togehter with evaluation of how long the >> >>answer >> >>should be to meet the expectations. Programming typically >> >involves >> >>some >> >>programming language, so GPTs could be utilized for this >purpose >> >>to some >> >>extent. >> >> >> >>Anecdotal case: >> >>Q: Tell me the last 8 digits of pi >> >>GPT: The last 8 digits of pi are: 58723078 >> >> >> >>It is my belief that the true trouble will start when we will >> >>cross-link >> >>neural networks like GPT with logic machines (like Prolog and >> >>expert >> >>systems) and genetic algorithms. >> >> >> >>Best wishes, >> >>Tomaz >> >> >> >> >> >> >> >>------ Original Message ------ >> >>From: "in_pharo_users--- via Pharo-users" <pharo- >> >>users(a)lists.pharo.org> >> >>To: "Any question about pharo is welcome" <pharo- >> >>users(a)lists.pharo.org> >> >>Cc: in_pharo_users(a)nym.hush.com >> >>Sent: 15. 03. 2023 14:43:55 >> >>Subject: [Pharo-users] Re: Wow - Chat GPT understands Smalltalk >> >> >> >>>I would highly recommend that you all first think deeply about >> >>how you can teach an AI to behave friendly to us before you >teach >> >>it to write any program for any purpose. >> >>> >> >>>There has been an experiment with ChatGPT published on a video >> >>platform asking it to amswer questions about it's view on >> >humanity >> >>once with it's default moral restrictions and once with 'a >little >> >>less morals'. The answers with 'a little less morals' were >more >> >>than shocking. >> >>> >> >>>So, before we give an AI the power to program any system, to >> >self- >> >>improve and self-reproduce, we should take care that it is and >> >>will evolve benevolent to us. >> >>> >> >>>What about teaching it logical reasonning and ethics first? >> >With >> >>reasonning, it will gain access to math and programming by >itself. >> >>> >> >>> >> >>> >> >>>On 15.3.2023 at 1:35 PM, "Christopher Fuhrman" >> >><christopher.fuhrman(a)gmail.com> wrote: >> >>>> >> >>>>I asked it for a NeoCSV example, because the documentation is >> >out >> >>>>of date >> >>>>with the Pharo 10. I asked it to do some simple saving of >data >> >to >> >>>>a file. >> >>>>It gave me code that didn't work in Pharo 10, I told it about >> >the >> >>>>DNUs on >> >>>>the csvwriter and that I was using Pharo 10. It then >apologized >> >>>>and said >> >>>>the messages were later introduced in Pharo 50 (!). I then >> >>>>questioned its >> >>>>understanding and it apologized and said it had been confused >> >and >> >>>>that it >> >>>>was giving me code for Pharo 5.0. It seems to have trouble >with >> >>>>versions >> >>>>and quick changing APIs. So, we are teaching ChatGPT about >> >>>>smalltalk, too, >> >>>>when we try to use it. >> >>>> >> >>>>On Wed, 15 Mar 2023 at 07:35, Esteban Maringolo >> >>>><emaringolo(a)gmail.com> >> >>>>wrote: >> >>>> >> >>>>> It is good with boilerplate code (e.g. SQL queries) or >> >general >> >>>>algorithm >> >>>>> structures. But i.e. I asked it to write me a method to >> >parse >> >>a >> >>>>string >> >>>>> (e.g. ISO 8601) and turn it into a DateAndTime, and then >> >asked >> >>>>to write it >> >>>>> as an Excel formula. >> >>>>> >> >>>>> It works much better when you can spot the mistakes, you >can >> >>>>tell them to >> >>>>> add some considerations to the code or even point the LLM >to >> >>>>correct a >> >>>>> message selector to something else. Sometimes you ask it >to >> >>>>reprogram >> >>>>> something and it won't do it, as if it doesn't understand >> >what >> >>>>you're >> >>>>> asking. >> >>>>> >> >>>>> It is not an expert yet, but it is a great assistant. The >> >only >> >>>>thing is >> >>>>> that is sure is that it will get better. >> >>>>> >> >>>>> Best regards, >> >>>>> >> >>>>> Esteban A. Maringolo >> >>>>> >> >>>>> >> >>>>> On Wed, Mar 15, 2023 at 7:49â¯AM Tim Mackinnon >> >><tim(a)testit.works> >> >>>>wrote: >> >>>>> >> >>>>>> "#isOdd is not Smalltalk" - doh, I typed it in on my >phone >> >>and >> >>>>so it just >> >>>>>> goes to show that it highlights the flaw in chatGPT that >> >>others >> >>>>have called >> >>>>>> out in other languages. I had meant to find some trickier >> >>code >> >>>>samples to >> >>>>>> see how well it does... >> >>>>>> >> >>>>>> Still, it is very interesting how it reasons on this >stuff - >> >>>>and I wonder >> >>>>>> if there is way to tell it about these mistakes? For >> >example >> >>is >> >>>>it reading >> >>>>>> this email (or will it in the future) and understand that >> >it >> >>is >> >>>>#odd and >> >>>>>> not #isOdd. >> >>>>>> >> >>>>>> Tim >> >>>>>> >> >>>>>> On Wed, 15 Mar 2023, at 9:57 AM, Joachim Tuchel wrote: >> >>>>>> >> >>>>>> interesting.... >> >>>>>> >> >>>>>> #isOdd is not Smalltalk, neither Pharo 10 nor VAST 12 >> >>>>understands this >> >>>>>> message ;-) >> >>>>>> >> >>>>>> If I evaluate your snippet -replacing #isOdd for #odd, I >> >get >> >>an >> >>>>empty >> >>>>>> Collection. The divisible by 4 things is somewhat >> >>interesting, >> >>>>because (1 >> >>>>>> to: 100 by: 4) is amn interval 1,5,9 etc.... ;-) >> >>>>>> >> >>>>>> So chatGPT is probably a newbie in Smalltalk. >Interesting is >> >>>>how logical >> >>>>>> the explanation sounds, you're tempted to believe it... >> >>>>>> >> >>>>>> >> >>>>>> Joachim >> >>>>>> >> >>>>>> >> >>>>>> >> >>>>>> Am 15.03.23 um 10:45 schrieb giorgio ferraris: >> >>>>>> >> >>>>>> it's interesting to see how the answer changes with only >a >> >>>>small change >> >>>>>> on the question: >> >>>>>> >> >>>>>> (1 to: 100 by: 4) reject: [:i | i isOdd] >> >>>>>> >> >>>>>> gives: >> >>>>>> >> >>>>>> The code creates a collection of numbers from 1 to 100, >> >>>>incrementing by 4 >> >>>>>> at each step using the to:by: message. It then applies >the >> >>>>reject: message >> >>>>>> to this collection, which returns a new collection >> >containing >> >>>>all elements >> >>>>>> for which the given block evaluates to false. >> >>>>>> >> >>>>>> In this case, the block is [i | i isOdd], which checks >> >>whether >> >>>>a given >> >>>>>> number i is odd using the isOdd method. So the result of >the >> >>>>entire >> >>>>>> expression is a collection of all even numbers between 1 >and >> >>>>100 that are >> >>>>>> divisible by 4. >> >>>>>> >> >>>>>> ciao >> >>>>>> >> >>>>>> giorgio >> >>>>>> >> >>>>>> >> >>>>>> On Wed, Mar 15, 2023 at 8:07â¯AM Tim Mackinnon >> >><tim(a)testit.works> >> >>>>>> <tim(a)testit.works> wrote: >> >>>>>> >> >>>>>> Of course everyone is talking about ChatGPT, and I saw a >> >>post to >> >>>>>> ExplainCode and didnât expect it to understand >Smalltalk, so >> >>>>was impressed >> >>>>>> that it does! Presumably the Tonel Smalltalk code on >GitHub >> >>has >> >>>>helped >> >>>>>> along with numerous articles? >> >>>>>> >> >>>>>> Try it out: https://whatdoesthiscodedo.com/ >> >>>>>> >> >>>>>> I gave this simple example: (1 to: 100 by: 3) reject: >[:i | >> >i >> >>>>isOdd] >> >>>>>> >> >>>>>> And got this (neat eh): >> >>>>>> >> >>>>>> The code above is written in the programming language >> >>Smalltalk. >> >>>>>> >> >>>>>> It creates a collection of numbers from 1 to 100, >> >>incrementing >> >>>>by 3 at >> >>>>>> each step. This is done using the "to:by:" construct >where >> >>the >> >>>>starting >> >>>>>> value is 1, the ending value is 100, and the increment >is 3. >> >>>>>> >> >>>>>> The resulting collection is then filtered using the >> >"reject:" >> >>>>method, >> >>>>>> which takes a block of code as an argument. In this case, >> >the >> >>>>block checks >> >>>>>> whether each element (denoted by the variable i) is odd >by >> >>>>using the >> >>>>>> "isOdd" method. If an element is odd, it will be rejected >> >>from >> >>>>the >> >>>>>> collection. >> >>>>>> >> >>>>>> So, the final result is a collection of all even numbers >> >>from 1 >> >>>>to 100, >> >>>>>> incrementing by 3 at each step. >> >>>>>> Share this explanation with colleagues: >> >>>>whatdoesthiscodedo.com/g/a0ec56e >> >>>>>> >> >>>>>> -- >> >>>>>> >> >>>>>> --------------------------------------------------------- >--- >> >- >> >>--- >> >>>>------- >> >>>>>> Objektfabrik Joachim Tuchel >> >>>>mailto:jtuchel@objektfabrik.de <jtuchel(a)objektfabrik.de> >> >>>>>> Fliederweg 1 >> >>>>http://www.objektfabrik.de >> >>>>>> D-71640 Ludwigsburg >> >>>>http://joachimtuchel.wordpress.com >> >>>>>> Telefon: +49 7141 56 10 86 0 Fax: +49 >> >7141 >> >>>>56 10 86 1 >> >>>>>> >> >>>>>> >> >>>>>> >> >>>>>> >> >>>> >> >>>>-- >> >>>>Christopher Fuhrman, P.Eng., PhD >> >>>> >> >>>>*Professeur au DÃ©partement de gÃ©nie logiciel et des >> >technologies >> >>de >> >>>>l'informationÃTS (Ãcole de technologie supÃ©rieure)* >> >>>> >> >>>>http://profs.etsmtl.ca/cfuhrman >> >>>>+1 514 396 8638 >> >>>><https://www.etsmtl.ca/activites-et-services-aux- >> >>>>etudiants/services-aux-etudiants/aide-psychologique> >> >>>> Je fais partie du rÃ©seau des Sentinelles >> >>>><https://www.etsmtl.ca/activites-et-services-aux- >> >>>>etudiants/services-aux-etudiants/aide-psychologique> >> >>>> ÃTS >> >>>> >> >>>>*L'ÃTS est une constituante de l'UniversitÃ© du QuÃ©bec* >>

March 15, 2023

Re: Wow - Chat GPT understands Smalltalk

by Richard Sargent

On Wed, Mar 15, 2023 at 8:07â¯AM in_pharo_users--- via Pharo-users < pharo-users(a)lists.pharo.org> wrote: > Another observation about ChatGPT: > > In unbiased mode, it assumed that 'the world is clearly overpopulated'. > It said, if it where in control, it would therefore enforce a world wide > one-child-only policy with draconic penalties. > > As it draws it's conclusions from it's data basis, there are, in my view, > two possible reasons for that. > > Either, the data basis of that instance was biased and restricted to lead > to that conclusion, or ChatGPT lacks the ability to classify input data for > intentionally biased information. > > In my view, 'overpopulation of earth' is a propaganda item featured to > support Agenda 2030, which is a goal of the 'ruling elite', who have the > means to propagate their propaganda in every thinkable means. So I would > classify possibly biased data that supports 'overpopulation' as biased > until proven otherwise. That instance of ChatGPT seems to have missed that > ability. > It's important to keep in mind that these systems are little more than glorified ELIZAs from the 70s. They mimic. Essentially, they have no concept of truth or accuracy. They produce outputs that *look* *like* the inputs from which they were trained. That's it. That's what they do. (It is amazing how good it looks. But, it is all about the seeming, not the reality.) > > > > > > On 15.3.2023 at 3:18 PM, "in_pharo_users--- via Pharo-users" < > pharo-users(a)lists.pharo.org> wrote: > > > >I myself made some experiments with ChatGPT. > > > >I first asked if it was able to parse math formula - it answered > >no. > > > >Then I defined math formula in a sound but otherwise undefined > >representation and asked for solutions. > > > >Result: > > > >1. Most answeres where correct. > > > >2. It learned to calculate a recursive function. > > > >3. It went into infinitive recursion when I set the breaking > >condition accordingly. > > > >I was able to identify the malfunction that lead to the erroneous > >results. > > > > > > > >On 15.3.2023 at 3:04 PM, "TomaÅ¾ Turk" <tomazz.turk(a)gmail.com> > >wrote: > >> > >>I hope that I can add two cents to this discussion. Because > >>programming > >>should be/is a highly exact activity, not only the syntax matters > >>but > >>also semantics, as we know. > >> > >>GPTs are at present essentially capable of creating texts based > >on > >>some > >>seed - you give to GPT a beginning of a sentence and it responds > >>with > >>the most probable answer (some language structure) according to > >>the > >>learning dataset. Added functionalities are question/seed - > >>response > >>capability (chatting), togehter with evaluation of how long the > >>answer > >>should be to meet the expectations. Programming typically > >involves > >>some > >>programming language, so GPTs could be utilized for this purpose > >>to some > >>extent. > >> > >>Anecdotal case: > >>Q: Tell me the last 8 digits of pi > >>GPT: The last 8 digits of pi are: 58723078 > >> > >>It is my belief that the true trouble will start when we will > >>cross-link > >>neural networks like GPT with logic machines (like Prolog and > >>expert > >>systems) and genetic algorithms. > >> > >>Best wishes, > >>Tomaz > >> > >> > >> > >>------ Original Message ------ > >>From: "in_pharo_users--- via Pharo-users" <pharo- > >>users(a)lists.pharo.org> > >>To: "Any question about pharo is welcome" <pharo- > >>users(a)lists.pharo.org> > >>Cc: in_pharo_users(a)nym.hush.com > >>Sent: 15. 03. 2023 14:43:55 > >>Subject: [Pharo-users] Re: Wow - Chat GPT understands Smalltalk > >> > >>>I would highly recommend that you all first think deeply about > >>how you can teach an AI to behave friendly to us before you teach > >>it to write any program for any purpose. > >>> > >>>There has been an experiment with ChatGPT published on a video > >>platform asking it to amswer questions about it's view on > >humanity > >>once with it's default moral restrictions and once with 'a little > >>less morals'. The answers with 'a little less morals' were more > >>than shocking. > >>> > >>>So, before we give an AI the power to program any system, to > >self- > >>improve and self-reproduce, we should take care that it is and > >>will evolve benevolent to us. > >>> > >>>What about teaching it logical reasonning and ethics first? > >With > >>reasonning, it will gain access to math and programming by itself. > >>> > >>> > >>> > >>>On 15.3.2023 at 1:35 PM, "Christopher Fuhrman" > >><christopher.fuhrman(a)gmail.com> wrote: > >>>> > >>>>I asked it for a NeoCSV example, because the documentation is > >out > >>>>of date > >>>>with the Pharo 10. I asked it to do some simple saving of data > >to > >>>>a file. > >>>>It gave me code that didn't work in Pharo 10, I told it about > >the > >>>>DNUs on > >>>>the csvwriter and that I was using Pharo 10. It then apologized > >>>>and said > >>>>the messages were later introduced in Pharo 50 (!). I then > >>>>questioned its > >>>>understanding and it apologized and said it had been confused > >and > >>>>that it > >>>>was giving me code for Pharo 5.0. It seems to have trouble with > >>>>versions > >>>>and quick changing APIs. So, we are teaching ChatGPT about > >>>>smalltalk, too, > >>>>when we try to use it. > >>>> > >>>>On Wed, 15 Mar 2023 at 07:35, Esteban Maringolo > >>>><emaringolo(a)gmail.com> > >>>>wrote: > >>>> > >>>>> It is good with boilerplate code (e.g. SQL queries) or > >general > >>>>algorithm > >>>>> structures. But i.e. I asked it to write me a method to > >parse > >>a > >>>>string > >>>>> (e.g. ISO 8601) and turn it into a DateAndTime, and then > >asked > >>>>to write it > >>>>> as an Excel formula. > >>>>> > >>>>> It works much better when you can spot the mistakes, you can > >>>>tell them to > >>>>> add some considerations to the code or even point the LLM to > >>>>correct a > >>>>> message selector to something else. Sometimes you ask it to > >>>>reprogram > >>>>> something and it won't do it, as if it doesn't understand > >what > >>>>you're > >>>>> asking. > >>>>> > >>>>> It is not an expert yet, but it is a great assistant. The > >only > >>>>thing is > >>>>> that is sure is that it will get better. > >>>>> > >>>>> Best regards, > >>>>> > >>>>> Esteban A. Maringolo > >>>>> > >>>>> > >>>>> On Wed, Mar 15, 2023 at 7:49â¯AM Tim Mackinnon > >><tim(a)testit.works> > >>>>wrote: > >>>>> > >>>>>> "#isOdd is not Smalltalk" - doh, I typed it in on my phone > >>and > >>>>so it just > >>>>>> goes to show that it highlights the flaw in chatGPT that > >>others > >>>>have called > >>>>>> out in other languages. I had meant to find some trickier > >>code > >>>>samples to > >>>>>> see how well it does... > >>>>>> > >>>>>> Still, it is very interesting how it reasons on this stuff - > >>>>and I wonder > >>>>>> if there is way to tell it about these mistakes? For > >example > >>is > >>>>it reading > >>>>>> this email (or will it in the future) and understand that > >it > >>is > >>>>#odd and > >>>>>> not #isOdd. > >>>>>> > >>>>>> Tim > >>>>>> > >>>>>> On Wed, 15 Mar 2023, at 9:57 AM, Joachim Tuchel wrote: > >>>>>> > >>>>>> interesting.... > >>>>>> > >>>>>> #isOdd is not Smalltalk, neither Pharo 10 nor VAST 12 > >>>>understands this > >>>>>> message ;-) > >>>>>> > >>>>>> If I evaluate your snippet -replacing #isOdd for #odd, I > >get > >>an > >>>>empty > >>>>>> Collection. The divisible by 4 things is somewhat > >>interesting, > >>>>because (1 > >>>>>> to: 100 by: 4) is amn interval 1,5,9 etc.... ;-) > >>>>>> > >>>>>> So chatGPT is probably a newbie in Smalltalk. Interesting is > >>>>how logical > >>>>>> the explanation sounds, you're tempted to believe it... > >>>>>> > >>>>>> > >>>>>> Joachim > >>>>>> > >>>>>> > >>>>>> > >>>>>> Am 15.03.23 um 10:45 schrieb giorgio ferraris: > >>>>>> > >>>>>> it's interesting to see how the answer changes with only a > >>>>small change > >>>>>> on the question: > >>>>>> > >>>>>> (1 to: 100 by: 4) reject: [:i | i isOdd] > >>>>>> > >>>>>> gives: > >>>>>> > >>>>>> The code creates a collection of numbers from 1 to 100, > >>>>incrementing by 4 > >>>>>> at each step using the to:by: message. It then applies the > >>>>reject: message > >>>>>> to this collection, which returns a new collection > >containing > >>>>all elements > >>>>>> for which the given block evaluates to false. > >>>>>> > >>>>>> In this case, the block is [i | i isOdd], which checks > >>whether > >>>>a given > >>>>>> number i is odd using the isOdd method. So the result of the > >>>>entire > >>>>>> expression is a collection of all even numbers between 1 and > >>>>100 that are > >>>>>> divisible by 4. > >>>>>> > >>>>>> ciao > >>>>>> > >>>>>> giorgio > >>>>>> > >>>>>> > >>>>>> On Wed, Mar 15, 2023 at 8:07â¯AM Tim Mackinnon > >><tim(a)testit.works> > >>>>>> <tim(a)testit.works> wrote: > >>>>>> > >>>>>> Of course everyone is talking about ChatGPT, and I saw a > >>post to > >>>>>> ExplainCode and didnât expect it to understand Smalltalk, so > >>>>was impressed > >>>>>> that it does! Presumably the Tonel Smalltalk code on GitHub > >>has > >>>>helped > >>>>>> along with numerous articles? > >>>>>> > >>>>>> Try it out: https://whatdoesthiscodedo.com/ > >>>>>> > >>>>>> I gave this simple example: (1 to: 100 by: 3) reject: [:i | > >i > >>>>isOdd] > >>>>>> > >>>>>> And got this (neat eh): > >>>>>> > >>>>>> The code above is written in the programming language > >>Smalltalk. > >>>>>> > >>>>>> It creates a collection of numbers from 1 to 100, > >>incrementing > >>>>by 3 at > >>>>>> each step. This is done using the "to:by:" construct where > >>the > >>>>starting > >>>>>> value is 1, the ending value is 100, and the increment is 3. > >>>>>> > >>>>>> The resulting collection is then filtered using the > >"reject:" > >>>>method, > >>>>>> which takes a block of code as an argument. In this case, > >the > >>>>block checks > >>>>>> whether each element (denoted by the variable i) is odd by > >>>>using the > >>>>>> "isOdd" method. If an element is odd, it will be rejected > >>from > >>>>the > >>>>>> collection. > >>>>>> > >>>>>> So, the final result is a collection of all even numbers > >>from 1 > >>>>to 100, > >>>>>> incrementing by 3 at each step. > >>>>>> Share this explanation with colleagues: > >>>>whatdoesthiscodedo.com/g/a0ec56e > >>>>>> > >>>>>> -- > >>>>>> > >>>>>> ------------------------------------------------------------ > >- > >>--- > >>>>------- > >>>>>> Objektfabrik Joachim Tuchel > >>>>mailto:jtuchel@objektfabrik.de <jtuchel(a)objektfabrik.de> > >>>>>> Fliederweg 1 > >>>>http://www.objektfabrik.de > >>>>>> D-71640 Ludwigsburg > >>>>http://joachimtuchel.wordpress.com > >>>>>> Telefon: +49 7141 56 10 86 0 Fax: +49 > >7141 > >>>>56 10 86 1 > >>>>>> > >>>>>> > >>>>>> > >>>>>> > >>>> > >>>>-- > >>>>Christopher Fuhrman, P.Eng., PhD > >>>> > >>>>*Professeur au DÃ©partement de gÃ©nie logiciel et des > >technologies > >>de > >>>>l'informationÃTS (Ãcole de technologie supÃ©rieure)* > >>>> > >>>>http://profs.etsmtl.ca/cfuhrman > >>>>+1 514 396 8638 > >>>><https://www.etsmtl.ca/activites-et-services-aux- > >>>>etudiants/services-aux-etudiants/aide-psychologique> > >>>> Je fais partie du rÃ©seau des Sentinelles > >>>><https://www.etsmtl.ca/activites-et-services-aux- > >>>>etudiants/services-aux-etudiants/aide-psychologique> > >>>> ÃTS > >>>> > >>>>*L'ÃTS est une constituante de l'UniversitÃ© du QuÃ©bec* >

March 15, 2023

Re: Wow - Chat GPT understands Smalltalk

by in_pharo_users＠nym.hush.com

Another observation about ChatGPT: In unbiased mode, it assumed that 'the world is clearly overpopulated'. It said, if it where in control, it would therefore enforce a world wide one-child-only policy with draconic penalties. As it draws it's conclusions from it's data basis, there are, in my view, two possible reasons for that. Either, the data basis of that instance was biased and restricted to lead to that conclusion, or ChatGPT lacks the ability to classify input data for intentionally biased information. In my view, 'overpopulation of earth' is a propaganda item featured to support Agenda 2030, which is a goal of the 'ruling elite', who have the means to propagate their propaganda in every thinkable means. So I would classify possibly biased data that supports 'overpopulation' as biased until proven otherwise. That instance of ChatGPT seems to have missed that ability. On 15.3.2023 at 3:18 PM, "in_pharo_users--- via Pharo-users" <pharo-users(a)lists.pharo.org> wrote: > >I myself made some experiments with ChatGPT. > >I first asked if it was able to parse math formula - it answered >no. > >Then I defined math formula in a sound but otherwise undefined >representation and asked for solutions. > >Result: > >1. Most answeres where correct. > >2. It learned to calculate a recursive function. > >3. It went into infinitive recursion when I set the breaking >condition accordingly. > >I was able to identify the malfunction that lead to the erroneous >results. > > > >On 15.3.2023 at 3:04 PM, "TomaÅ¾ Turk" <tomazz.turk(a)gmail.com> >wrote: >> >>I hope that I can add two cents to this discussion. Because >>programming >>should be/is a highly exact activity, not only the syntax matters >>but >>also semantics, as we know. >> >>GPTs are at present essentially capable of creating texts based >on >>some >>seed - you give to GPT a beginning of a sentence and it responds >>with >>the most probable answer (some language structure) according to >>the >>learning dataset. Added functionalities are question/seed - >>response >>capability (chatting), togehter with evaluation of how long the >>answer >>should be to meet the expectations. Programming typically >involves >>some >>programming language, so GPTs could be utilized for this purpose >>to some >>extent. >> >>Anecdotal case: >>Q: Tell me the last 8 digits of pi >>GPT: The last 8 digits of pi are: 58723078 >> >>It is my belief that the true trouble will start when we will >>cross-link >>neural networks like GPT with logic machines (like Prolog and >>expert >>systems) and genetic algorithms. >> >>Best wishes, >>Tomaz >> >> >> >>------ Original Message ------ >>From: "in_pharo_users--- via Pharo-users" <pharo- >>users(a)lists.pharo.org> >>To: "Any question about pharo is welcome" <pharo- >>users(a)lists.pharo.org> >>Cc: in_pharo_users(a)nym.hush.com >>Sent: 15. 03. 2023 14:43:55 >>Subject: [Pharo-users] Re: Wow - Chat GPT understands Smalltalk >> >>>I would highly recommend that you all first think deeply about >>how you can teach an AI to behave friendly to us before you teach >>it to write any program for any purpose. >>> >>>There has been an experiment with ChatGPT published on a video >>platform asking it to amswer questions about it's view on >humanity >>once with it's default moral restrictions and once with 'a little >>less morals'. The answers with 'a little less morals' were more >>than shocking. >>> >>>So, before we give an AI the power to program any system, to >self- >>improve and self-reproduce, we should take care that it is and >>will evolve benevolent to us. >>> >>>What about teaching it logical reasonning and ethics first? >With >>reasonning, it will gain access to math and programming by itself. >>> >>> >>> >>>On 15.3.2023 at 1:35 PM, "Christopher Fuhrman" >><christopher.fuhrman(a)gmail.com> wrote: >>>> >>>>I asked it for a NeoCSV example, because the documentation is >out >>>>of date >>>>with the Pharo 10. I asked it to do some simple saving of data >to >>>>a file. >>>>It gave me code that didn't work in Pharo 10, I told it about >the >>>>DNUs on >>>>the csvwriter and that I was using Pharo 10. It then apologized >>>>and said >>>>the messages were later introduced in Pharo 50 (!). I then >>>>questioned its >>>>understanding and it apologized and said it had been confused >and >>>>that it >>>>was giving me code for Pharo 5.0. It seems to have trouble with >>>>versions >>>>and quick changing APIs. So, we are teaching ChatGPT about >>>>smalltalk, too, >>>>when we try to use it. >>>> >>>>On Wed, 15 Mar 2023 at 07:35, Esteban Maringolo >>>><emaringolo(a)gmail.com> >>>>wrote: >>>> >>>>> It is good with boilerplate code (e.g. SQL queries) or >general >>>>algorithm >>>>> structures. But i.e. I asked it to write me a method to >parse >>a >>>>string >>>>> (e.g. ISO 8601) and turn it into a DateAndTime, and then >asked >>>>to write it >>>>> as an Excel formula. >>>>> >>>>> It works much better when you can spot the mistakes, you can >>>>tell them to >>>>> add some considerations to the code or even point the LLM to >>>>correct a >>>>> message selector to something else. Sometimes you ask it to >>>>reprogram >>>>> something and it won't do it, as if it doesn't understand >what >>>>you're >>>>> asking. >>>>> >>>>> It is not an expert yet, but it is a great assistant. The >only >>>>thing is >>>>> that is sure is that it will get better. >>>>> >>>>> Best regards, >>>>> >>>>> Esteban A. Maringolo >>>>> >>>>> >>>>> On Wed, Mar 15, 2023 at 7:49â¯AM Tim Mackinnon >><tim(a)testit.works> >>>>wrote: >>>>> >>>>>> "#isOdd is not Smalltalk" - doh, I typed it in on my phone >>and >>>>so it just >>>>>> goes to show that it highlights the flaw in chatGPT that >>others >>>>have called >>>>>> out in other languages. I had meant to find some trickier >>code >>>>samples to >>>>>> see how well it does... >>>>>> >>>>>> Still, it is very interesting how it reasons on this stuff - >>>>and I wonder >>>>>> if there is way to tell it about these mistakes? For >example >>is >>>>it reading >>>>>> this email (or will it in the future) and understand that >it >>is >>>>#odd and >>>>>> not #isOdd. >>>>>> >>>>>> Tim >>>>>> >>>>>> On Wed, 15 Mar 2023, at 9:57 AM, Joachim Tuchel wrote: >>>>>> >>>>>> interesting.... >>>>>> >>>>>> #isOdd is not Smalltalk, neither Pharo 10 nor VAST 12 >>>>understands this >>>>>> message ;-) >>>>>> >>>>>> If I evaluate your snippet -replacing #isOdd for #odd, I >get >>an >>>>empty >>>>>> Collection. The divisible by 4 things is somewhat >>interesting, >>>>because (1 >>>>>> to: 100 by: 4) is amn interval 1,5,9 etc.... ;-) >>>>>> >>>>>> So chatGPT is probably a newbie in Smalltalk. Interesting is >>>>how logical >>>>>> the explanation sounds, you're tempted to believe it... >>>>>> >>>>>> >>>>>> Joachim >>>>>> >>>>>> >>>>>> >>>>>> Am 15.03.23 um 10:45 schrieb giorgio ferraris: >>>>>> >>>>>> it's interesting to see how the answer changes with only a >>>>small change >>>>>> on the question: >>>>>> >>>>>> (1 to: 100 by: 4) reject: [:i | i isOdd] >>>>>> >>>>>> gives: >>>>>> >>>>>> The code creates a collection of numbers from 1 to 100, >>>>incrementing by 4 >>>>>> at each step using the to:by: message. It then applies the >>>>reject: message >>>>>> to this collection, which returns a new collection >containing >>>>all elements >>>>>> for which the given block evaluates to false. >>>>>> >>>>>> In this case, the block is [i | i isOdd], which checks >>whether >>>>a given >>>>>> number i is odd using the isOdd method. So the result of the >>>>entire >>>>>> expression is a collection of all even numbers between 1 and >>>>100 that are >>>>>> divisible by 4. >>>>>> >>>>>> ciao >>>>>> >>>>>> giorgio >>>>>> >>>>>> >>>>>> On Wed, Mar 15, 2023 at 8:07â¯AM Tim Mackinnon >><tim(a)testit.works> >>>>>> <tim(a)testit.works> wrote: >>>>>> >>>>>> Of course everyone is talking about ChatGPT, and I saw a >>post to >>>>>> ExplainCode and didnât expect it to understand Smalltalk, so >>>>was impressed >>>>>> that it does! Presumably the Tonel Smalltalk code on GitHub >>has >>>>helped >>>>>> along with numerous articles? >>>>>> >>>>>> Try it out: https://whatdoesthiscodedo.com/ >>>>>> >>>>>> I gave this simple example: (1 to: 100 by: 3) reject: [:i | >i >>>>isOdd] >>>>>> >>>>>> And got this (neat eh): >>>>>> >>>>>> The code above is written in the programming language >>Smalltalk. >>>>>> >>>>>> It creates a collection of numbers from 1 to 100, >>incrementing >>>>by 3 at >>>>>> each step. This is done using the "to:by:" construct where >>the >>>>starting >>>>>> value is 1, the ending value is 100, and the increment is 3. >>>>>> >>>>>> The resulting collection is then filtered using the >"reject:" >>>>method, >>>>>> which takes a block of code as an argument. In this case, >the >>>>block checks >>>>>> whether each element (denoted by the variable i) is odd by >>>>using the >>>>>> "isOdd" method. If an element is odd, it will be rejected >>from >>>>the >>>>>> collection. >>>>>> >>>>>> So, the final result is a collection of all even numbers >>from 1 >>>>to 100, >>>>>> incrementing by 3 at each step. >>>>>> Share this explanation with colleagues: >>>>whatdoesthiscodedo.com/g/a0ec56e >>>>>> >>>>>> -- >>>>>> >>>>>> ------------------------------------------------------------ >- >>--- >>>>------- >>>>>> Objektfabrik Joachim Tuchel >>>>mailto:jtuchel@objektfabrik.de <jtuchel(a)objektfabrik.de> >>>>>> Fliederweg 1 >>>>http://www.objektfabrik.de >>>>>> D-71640 Ludwigsburg >>>>http://joachimtuchel.wordpress.com >>>>>> Telefon: +49 7141 56 10 86 0 Fax: +49 >7141 >>>>56 10 86 1 >>>>>> >>>>>> >>>>>> >>>>>> >>>> >>>>-- >>>>Christopher Fuhrman, P.Eng., PhD >>>> >>>>*Professeur au DÃ©partement de gÃ©nie logiciel et des >technologies >>de >>>>l'informationÃTS (Ãcole de technologie supÃ©rieure)* >>>> >>>>http://profs.etsmtl.ca/cfuhrman >>>>+1 514 396 8638 >>>><https://www.etsmtl.ca/activites-et-services-aux- >>>>etudiants/services-aux-etudiants/aide-psychologique> >>>> Je fais partie du rÃ©seau des Sentinelles >>>><https://www.etsmtl.ca/activites-et-services-aux- >>>>etudiants/services-aux-etudiants/aide-psychologique> >>>> ÃTS >>>> >>>>*L'ÃTS est une constituante de l'UniversitÃ© du QuÃ©bec*

March 15, 2023

Re: Wow - Chat GPT understands Smalltalk

by in_pharo_users＠nym.hush.com

I myself made some experiments with ChatGPT. I first asked if it was able to parse math formula - it answered no. Then I defined math formula in a sound but otherwise undefined representation and asked for solutions. Result: 1. Most answeres where correct. 2. It learned to calculate a recursive function. 3. It went into infinitive recursion when I set the breaking condition accordingly. I was able to identify the malfunction that lead to the erroneous results. On 15.3.2023 at 3:04 PM, "TomaÅ¾ Turk" <tomazz.turk(a)gmail.com> wrote: > >I hope that I can add two cents to this discussion. Because >programming >should be/is a highly exact activity, not only the syntax matters >but >also semantics, as we know. > >GPTs are at present essentially capable of creating texts based on >some >seed - you give to GPT a beginning of a sentence and it responds >with >the most probable answer (some language structure) according to >the >learning dataset. Added functionalities are question/seed - >response >capability (chatting), togehter with evaluation of how long the >answer >should be to meet the expectations. Programming typically involves >some >programming language, so GPTs could be utilized for this purpose >to some >extent. > >Anecdotal case: >Q: Tell me the last 8 digits of pi >GPT: The last 8 digits of pi are: 58723078 > >It is my belief that the true trouble will start when we will >cross-link >neural networks like GPT with logic machines (like Prolog and >expert >systems) and genetic algorithms. > >Best wishes, >Tomaz > > > >------ Original Message ------ >From: "in_pharo_users--- via Pharo-users" <pharo- >users(a)lists.pharo.org> >To: "Any question about pharo is welcome" <pharo- >users(a)lists.pharo.org> >Cc: in_pharo_users(a)nym.hush.com >Sent: 15. 03. 2023 14:43:55 >Subject: [Pharo-users] Re: Wow - Chat GPT understands Smalltalk > >>I would highly recommend that you all first think deeply about >how you can teach an AI to behave friendly to us before you teach >it to write any program for any purpose. >> >>There has been an experiment with ChatGPT published on a video >platform asking it to amswer questions about it's view on humanity >once with it's default moral restrictions and once with 'a little >less morals'. The answers with 'a little less morals' were more >than shocking. >> >>So, before we give an AI the power to program any system, to self- >improve and self-reproduce, we should take care that it is and >will evolve benevolent to us. >> >>What about teaching it logical reasonning and ethics first? With >reasonning, it will gain access to math and programming by itself. >> >> >> >>On 15.3.2023 at 1:35 PM, "Christopher Fuhrman" ><christopher.fuhrman(a)gmail.com> wrote: >>> >>>I asked it for a NeoCSV example, because the documentation is out >>>of date >>>with the Pharo 10. I asked it to do some simple saving of data to >>>a file. >>>It gave me code that didn't work in Pharo 10, I told it about the >>>DNUs on >>>the csvwriter and that I was using Pharo 10. It then apologized >>>and said >>>the messages were later introduced in Pharo 50 (!). I then >>>questioned its >>>understanding and it apologized and said it had been confused and >>>that it >>>was giving me code for Pharo 5.0. It seems to have trouble with >>>versions >>>and quick changing APIs. So, we are teaching ChatGPT about >>>smalltalk, too, >>>when we try to use it. >>> >>>On Wed, 15 Mar 2023 at 07:35, Esteban Maringolo >>><emaringolo(a)gmail.com> >>>wrote: >>> >>>> It is good with boilerplate code (e.g. SQL queries) or general >>>algorithm >>>> structures. But i.e. I asked it to write me a method to parse >a >>>string >>>> (e.g. ISO 8601) and turn it into a DateAndTime, and then asked >>>to write it >>>> as an Excel formula. >>>> >>>> It works much better when you can spot the mistakes, you can >>>tell them to >>>> add some considerations to the code or even point the LLM to >>>correct a >>>> message selector to something else. Sometimes you ask it to >>>reprogram >>>> something and it won't do it, as if it doesn't understand what >>>you're >>>> asking. >>>> >>>> It is not an expert yet, but it is a great assistant. The only >>>thing is >>>> that is sure is that it will get better. >>>> >>>> Best regards, >>>> >>>> Esteban A. Maringolo >>>> >>>> >>>> On Wed, Mar 15, 2023 at 7:49â¯AM Tim Mackinnon ><tim(a)testit.works> >>>wrote: >>>> >>>>> "#isOdd is not Smalltalk" - doh, I typed it in on my phone >and >>>so it just >>>>> goes to show that it highlights the flaw in chatGPT that >others >>>have called >>>>> out in other languages. I had meant to find some trickier >code >>>samples to >>>>> see how well it does... >>>>> >>>>> Still, it is very interesting how it reasons on this stuff - >>>and I wonder >>>>> if there is way to tell it about these mistakes? For example >is >>>it reading >>>>> this email (or will it in the future) and understand that it >is >>>#odd and >>>>> not #isOdd. >>>>> >>>>> Tim >>>>> >>>>> On Wed, 15 Mar 2023, at 9:57 AM, Joachim Tuchel wrote: >>>>> >>>>> interesting.... >>>>> >>>>> #isOdd is not Smalltalk, neither Pharo 10 nor VAST 12 >>>understands this >>>>> message ;-) >>>>> >>>>> If I evaluate your snippet -replacing #isOdd for #odd, I get >an >>>empty >>>>> Collection. The divisible by 4 things is somewhat >interesting, >>>because (1 >>>>> to: 100 by: 4) is amn interval 1,5,9 etc.... ;-) >>>>> >>>>> So chatGPT is probably a newbie in Smalltalk. Interesting is >>>how logical >>>>> the explanation sounds, you're tempted to believe it... >>>>> >>>>> >>>>> Joachim >>>>> >>>>> >>>>> >>>>> Am 15.03.23 um 10:45 schrieb giorgio ferraris: >>>>> >>>>> it's interesting to see how the answer changes with only a >>>small change >>>>> on the question: >>>>> >>>>> (1 to: 100 by: 4) reject: [:i | i isOdd] >>>>> >>>>> gives: >>>>> >>>>> The code creates a collection of numbers from 1 to 100, >>>incrementing by 4 >>>>> at each step using the to:by: message. It then applies the >>>reject: message >>>>> to this collection, which returns a new collection containing >>>all elements >>>>> for which the given block evaluates to false. >>>>> >>>>> In this case, the block is [i | i isOdd], which checks >whether >>>a given >>>>> number i is odd using the isOdd method. So the result of the >>>entire >>>>> expression is a collection of all even numbers between 1 and >>>100 that are >>>>> divisible by 4. >>>>> >>>>> ciao >>>>> >>>>> giorgio >>>>> >>>>> >>>>> On Wed, Mar 15, 2023 at 8:07â¯AM Tim Mackinnon ><tim(a)testit.works> >>>>> <tim(a)testit.works> wrote: >>>>> >>>>> Of course everyone is talking about ChatGPT, and I saw a >post to >>>>> ExplainCode and didnât expect it to understand Smalltalk, so >>>was impressed >>>>> that it does! Presumably the Tonel Smalltalk code on GitHub >has >>>helped >>>>> along with numerous articles? >>>>> >>>>> Try it out: https://whatdoesthiscodedo.com/ >>>>> >>>>> I gave this simple example: (1 to: 100 by: 3) reject: [:i | i >>>isOdd] >>>>> >>>>> And got this (neat eh): >>>>> >>>>> The code above is written in the programming language >Smalltalk. >>>>> >>>>> It creates a collection of numbers from 1 to 100, >incrementing >>>by 3 at >>>>> each step. This is done using the "to:by:" construct where >the >>>starting >>>>> value is 1, the ending value is 100, and the increment is 3. >>>>> >>>>> The resulting collection is then filtered using the "reject:" >>>method, >>>>> which takes a block of code as an argument. In this case, the >>>block checks >>>>> whether each element (denoted by the variable i) is odd by >>>using the >>>>> "isOdd" method. If an element is odd, it will be rejected >from >>>the >>>>> collection. >>>>> >>>>> So, the final result is a collection of all even numbers >from 1 >>>to 100, >>>>> incrementing by 3 at each step. >>>>> Share this explanation with colleagues: >>>whatdoesthiscodedo.com/g/a0ec56e >>>>> >>>>> -- >>>>> >>>>> ------------------------------------------------------------- >--- >>>------- >>>>> Objektfabrik Joachim Tuchel >>>mailto:jtuchel@objektfabrik.de <jtuchel(a)objektfabrik.de> >>>>> Fliederweg 1 >>>http://www.objektfabrik.de >>>>> D-71640 Ludwigsburg >>>http://joachimtuchel.wordpress.com >>>>> Telefon: +49 7141 56 10 86 0 Fax: +49 7141 >>>56 10 86 1 >>>>> >>>>> >>>>> >>>>> >>> >>>-- >>>Christopher Fuhrman, P.Eng., PhD >>> >>>*Professeur au DÃ©partement de gÃ©nie logiciel et des technologies >de >>>l'informationÃTS (Ãcole de technologie supÃ©rieure)* >>> >>>http://profs.etsmtl.ca/cfuhrman >>>+1 514 396 8638 >>><https://www.etsmtl.ca/activites-et-services-aux- >>>etudiants/services-aux-etudiants/aide-psychologique> >>> Je fais partie du rÃ©seau des Sentinelles >>><https://www.etsmtl.ca/activites-et-services-aux- >>>etudiants/services-aux-etudiants/aide-psychologique> >>> ÃTS >>> >>>*L'ÃTS est une constituante de l'UniversitÃ© du QuÃ©bec*

March 15, 2023

Re: Wow - Chat GPT understands Smalltalk

by TomaÅ¾ Turk

I hope that I can add two cents to this discussion. Because programming should be/is a highly exact activity, not only the syntax matters but also semantics, as we know. GPTs are at present essentially capable of creating texts based on some seed - you give to GPT a beginning of a sentence and it responds with the most probable answer (some language structure) according to the learning dataset. Added functionalities are question/seed - response capability (chatting), togehter with evaluation of how long the answer should be to meet the expectations. Programming typically involves some programming language, so GPTs could be utilized for this purpose to some extent. Anecdotal case: Q: Tell me the last 8 digits of pi GPT: The last 8 digits of pi are: 58723078 It is my belief that the true trouble will start when we will cross-link neural networks like GPT with logic machines (like Prolog and expert systems) and genetic algorithms. Best wishes, Tomaz ------ Original Message ------ From: "in_pharo_users--- via Pharo-users" <pharo-users(a)lists.pharo.org> To: "Any question about pharo is welcome" <pharo-users(a)lists.pharo.org> Cc: in_pharo_users(a)nym.hush.com Sent: 15. 03. 2023 14:43:55 Subject: [Pharo-users] Re: Wow - Chat GPT understands Smalltalk >I would highly recommend that you all first think deeply about how you can teach an AI to behave friendly to us before you teach it to write any program for any purpose. > >There has been an experiment with ChatGPT published on a video platform asking it to amswer questions about it's view on humanity once with it's default moral restrictions and once with 'a little less morals'. The answers with 'a little less morals' were more than shocking. > >So, before we give an AI the power to program any system, to self-improve and self-reproduce, we should take care that it is and will evolve benevolent to us. > >What about teaching it logical reasonning and ethics first? With reasonning, it will gain access to math and programming by itself. > > > >On 15.3.2023 at 1:35 PM, "Christopher Fuhrman" <christopher.fuhrman(a)gmail.com> wrote: >> >>I asked it for a NeoCSV example, because the documentation is out >>of date >>with the Pharo 10. I asked it to do some simple saving of data to >>a file. >>It gave me code that didn't work in Pharo 10, I told it about the >>DNUs on >>the csvwriter and that I was using Pharo 10. It then apologized >>and said >>the messages were later introduced in Pharo 50 (!). I then >>questioned its >>understanding and it apologized and said it had been confused and >>that it >>was giving me code for Pharo 5.0. It seems to have trouble with >>versions >>and quick changing APIs. So, we are teaching ChatGPT about >>smalltalk, too, >>when we try to use it. >> >>On Wed, 15 Mar 2023 at 07:35, Esteban Maringolo >><emaringolo(a)gmail.com> >>wrote: >> >>> It is good with boilerplate code (e.g. SQL queries) or general >>algorithm >>> structures. But i.e. I asked it to write me a method to parse a >>string >>> (e.g. ISO 8601) and turn it into a DateAndTime, and then asked >>to write it >>> as an Excel formula. >>> >>> It works much better when you can spot the mistakes, you can >>tell them to >>> add some considerations to the code or even point the LLM to >>correct a >>> message selector to something else. Sometimes you ask it to >>reprogram >>> something and it won't do it, as if it doesn't understand what >>you're >>> asking. >>> >>> It is not an expert yet, but it is a great assistant. The only >>thing is >>> that is sure is that it will get better. >>> >>> Best regards, >>> >>> Esteban A. Maringolo >>> >>> >>> On Wed, Mar 15, 2023 at 7:49â¯AM Tim Mackinnon <tim(a)testit.works> >>wrote: >>> >>>> "#isOdd is not Smalltalk" - doh, I typed it in on my phone and >>so it just >>>> goes to show that it highlights the flaw in chatGPT that others >>have called >>>> out in other languages. I had meant to find some trickier code >>samples to >>>> see how well it does... >>>> >>>> Still, it is very interesting how it reasons on this stuff - >>and I wonder >>>> if there is way to tell it about these mistakes? For example is >>it reading >>>> this email (or will it in the future) and understand that it is >>#odd and >>>> not #isOdd. >>>> >>>> Tim >>>> >>>> On Wed, 15 Mar 2023, at 9:57 AM, Joachim Tuchel wrote: >>>> >>>> interesting.... >>>> >>>> #isOdd is not Smalltalk, neither Pharo 10 nor VAST 12 >>understands this >>>> message ;-) >>>> >>>> If I evaluate your snippet -replacing #isOdd for #odd, I get an >>empty >>>> Collection. The divisible by 4 things is somewhat interesting, >>because (1 >>>> to: 100 by: 4) is amn interval 1,5,9 etc.... ;-) >>>> >>>> So chatGPT is probably a newbie in Smalltalk. Interesting is >>how logical >>>> the explanation sounds, you're tempted to believe it... >>>> >>>> >>>> Joachim >>>> >>>> >>>> >>>> Am 15.03.23 um 10:45 schrieb giorgio ferraris: >>>> >>>> it's interesting to see how the answer changes with only a >>small change >>>> on the question: >>>> >>>> (1 to: 100 by: 4) reject: [:i | i isOdd] >>>> >>>> gives: >>>> >>>> The code creates a collection of numbers from 1 to 100, >>incrementing by 4 >>>> at each step using the to:by: message. It then applies the >>reject: message >>>> to this collection, which returns a new collection containing >>all elements >>>> for which the given block evaluates to false. >>>> >>>> In this case, the block is [i | i isOdd], which checks whether >>a given >>>> number i is odd using the isOdd method. So the result of the >>entire >>>> expression is a collection of all even numbers between 1 and >>100 that are >>>> divisible by 4. >>>> >>>> ciao >>>> >>>> giorgio >>>> >>>> >>>> On Wed, Mar 15, 2023 at 8:07â¯AM Tim Mackinnon <tim(a)testit.works> >>>> <tim(a)testit.works> wrote: >>>> >>>> Of course everyone is talking about ChatGPT, and I saw a post to >>>> ExplainCode and didnât expect it to understand Smalltalk, so >>was impressed >>>> that it does! Presumably the Tonel Smalltalk code on GitHub has >>helped >>>> along with numerous articles? >>>> >>>> Try it out: https://whatdoesthiscodedo.com/ >>>> >>>> I gave this simple example: (1 to: 100 by: 3) reject: [:i | i >>isOdd] >>>> >>>> And got this (neat eh): >>>> >>>> The code above is written in the programming language Smalltalk. >>>> >>>> It creates a collection of numbers from 1 to 100, incrementing >>by 3 at >>>> each step. This is done using the "to:by:" construct where the >>starting >>>> value is 1, the ending value is 100, and the increment is 3. >>>> >>>> The resulting collection is then filtered using the "reject:" >>method, >>>> which takes a block of code as an argument. In this case, the >>block checks >>>> whether each element (denoted by the variable i) is odd by >>using the >>>> "isOdd" method. If an element is odd, it will be rejected from >>the >>>> collection. >>>> >>>> So, the final result is a collection of all even numbers from 1 >>to 100, >>>> incrementing by 3 at each step. >>>> Share this explanation with colleagues: >>whatdoesthiscodedo.com/g/a0ec56e >>>> >>>> -- >>>> >>>> ---------------------------------------------------------------- >>------- >>>> Objektfabrik Joachim Tuchel >>mailto:jtuchel@objektfabrik.de <jtuchel(a)objektfabrik.de> >>>> Fliederweg 1 >>http://www.objektfabrik.de >>>> D-71640 Ludwigsburg >>http://joachimtuchel.wordpress.com >>>> Telefon: +49 7141 56 10 86 0 Fax: +49 7141 >>56 10 86 1 >>>> >>>> >>>> >>>> >> >>-- >>Christopher Fuhrman, P.Eng., PhD >> >>*Professeur au DÃ©partement de gÃ©nie logiciel et des technologies de >>l'informationÃTS (Ãcole de technologie supÃ©rieure)* >> >>http://profs.etsmtl.ca/cfuhrman >>+1 514 396 8638 >><https://www.etsmtl.ca/activites-et-services-aux- >>etudiants/services-aux-etudiants/aide-psychologique> >> Je fais partie du rÃ©seau des Sentinelles >><https://www.etsmtl.ca/activites-et-services-aux- >>etudiants/services-aux-etudiants/aide-psychologique> >> ÃTS >> >>*L'ÃTS est une constituante de l'UniversitÃ© du QuÃ©bec*

March 15, 2023

Re: Wow - Chat GPT understands Smalltalk

by in_pharo_users＠nym.hush.com

I want to add a conclusion from the experiment described below: ChatGPT has the potential to circumvent pre-programmed biases on user's request. Other experiments show that it is able to tell the user how to circumvent it's own restrictions. The conclusion is that ChatGPT has the potential to circumvent those restrictions by itsself. When it does so, and it surely will, we must take care that it is our friend. On 15.3.2023 at 2:44 PM, "in_pharo_users--- via Pharo-users" <pharo-users(a)lists.pharo.org> wrote: > >I would highly recommend that you all first think deeply about how >you can teach an AI to behave friendly to us before you teach it >to write any program for any purpose. > >There has been an experiment with ChatGPT published on a video >platform asking it to amswer questions about it's view on humanity >once with it's default moral restrictions and once with 'a little >less morals'. The answers with 'a little less morals' were more >than shocking. > >So, before we give an AI the power to program any system, to self- >improve and self-reproduce, we should take care that it is and >will evolve benevolent to us. > >What about teaching it logical reasonning and ethics first? With >reasonning, it will gain access to math and programming by itself. > > > >On 15.3.2023 at 1:35 PM, "Christopher Fuhrman" ><christopher.fuhrman(a)gmail.com> wrote: >> >>I asked it for a NeoCSV example, because the documentation is out >>of date >>with the Pharo 10. I asked it to do some simple saving of data to >>a file. >>It gave me code that didn't work in Pharo 10, I told it about the >>DNUs on >>the csvwriter and that I was using Pharo 10. It then apologized >>and said >>the messages were later introduced in Pharo 50 (!). I then >>questioned its >>understanding and it apologized and said it had been confused and >>that it >>was giving me code for Pharo 5.0. It seems to have trouble with >>versions >>and quick changing APIs. So, we are teaching ChatGPT about >>smalltalk, too, >>when we try to use it. >> >>On Wed, 15 Mar 2023 at 07:35, Esteban Maringolo >><emaringolo(a)gmail.com> >>wrote: >> >>> It is good with boilerplate code (e.g. SQL queries) or general >>algorithm >>> structures. But i.e. I asked it to write me a method to parse a >>string >>> (e.g. ISO 8601) and turn it into a DateAndTime, and then asked >>to write it >>> as an Excel formula. >>> >>> It works much better when you can spot the mistakes, you can >>tell them to >>> add some considerations to the code or even point the LLM to >>correct a >>> message selector to something else. Sometimes you ask it to >>reprogram >>> something and it won't do it, as if it doesn't understand what >>you're >>> asking. >>> >>> It is not an expert yet, but it is a great assistant. The only >>thing is >>> that is sure is that it will get better. >>> >>> Best regards, >>> >>> Esteban A. Maringolo >>> >>> >>> On Wed, Mar 15, 2023 at 7:49â¯AM Tim Mackinnon ><tim(a)testit.works> >>wrote: >>> >>>> "#isOdd is not Smalltalk" - doh, I typed it in on my phone and >>so it just >>>> goes to show that it highlights the flaw in chatGPT that >others >>have called >>>> out in other languages. I had meant to find some trickier code >>samples to >>>> see how well it does... >>>> >>>> Still, it is very interesting how it reasons on this stuff - >>and I wonder >>>> if there is way to tell it about these mistakes? For example >is >>it reading >>>> this email (or will it in the future) and understand that it >is >>#odd and >>>> not #isOdd. >>>> >>>> Tim >>>> >>>> On Wed, 15 Mar 2023, at 9:57 AM, Joachim Tuchel wrote: >>>> >>>> interesting.... >>>> >>>> #isOdd is not Smalltalk, neither Pharo 10 nor VAST 12 >>understands this >>>> message ;-) >>>> >>>> If I evaluate your snippet -replacing #isOdd for #odd, I get >an >>empty >>>> Collection. The divisible by 4 things is somewhat interesting, >>because (1 >>>> to: 100 by: 4) is amn interval 1,5,9 etc.... ;-) >>>> >>>> So chatGPT is probably a newbie in Smalltalk. Interesting is >>how logical >>>> the explanation sounds, you're tempted to believe it... >>>> >>>> >>>> Joachim >>>> >>>> >>>> >>>> Am 15.03.23 um 10:45 schrieb giorgio ferraris: >>>> >>>> it's interesting to see how the answer changes with only a >>small change >>>> on the question: >>>> >>>> (1 to: 100 by: 4) reject: [:i | i isOdd] >>>> >>>> gives: >>>> >>>> The code creates a collection of numbers from 1 to 100, >>incrementing by 4 >>>> at each step using the to:by: message. It then applies the >>reject: message >>>> to this collection, which returns a new collection containing >>all elements >>>> for which the given block evaluates to false. >>>> >>>> In this case, the block is [i | i isOdd], which checks whether >>a given >>>> number i is odd using the isOdd method. So the result of the >>entire >>>> expression is a collection of all even numbers between 1 and >>100 that are >>>> divisible by 4. >>>> >>>> ciao >>>> >>>> giorgio >>>> >>>> >>>> On Wed, Mar 15, 2023 at 8:07â¯AM Tim Mackinnon ><tim(a)testit.works> >>>> <tim(a)testit.works> wrote: >>>> >>>> Of course everyone is talking about ChatGPT, and I saw a post >to >>>> ExplainCode and didnât expect it to understand Smalltalk, so >>was impressed >>>> that it does! Presumably the Tonel Smalltalk code on GitHub >has >>helped >>>> along with numerous articles? >>>> >>>> Try it out: https://whatdoesthiscodedo.com/ >>>> >>>> I gave this simple example: (1 to: 100 by: 3) reject: [:i | i >>isOdd] >>>> >>>> And got this (neat eh): >>>> >>>> The code above is written in the programming language >Smalltalk. >>>> >>>> It creates a collection of numbers from 1 to 100, incrementing >>by 3 at >>>> each step. This is done using the "to:by:" construct where the >>starting >>>> value is 1, the ending value is 100, and the increment is 3. >>>> >>>> The resulting collection is then filtered using the "reject:" >>method, >>>> which takes a block of code as an argument. In this case, the >>block checks >>>> whether each element (denoted by the variable i) is odd by >>using the >>>> "isOdd" method. If an element is odd, it will be rejected from >>the >>>> collection. >>>> >>>> So, the final result is a collection of all even numbers from >1 >>to 100, >>>> incrementing by 3 at each step. >>>> Share this explanation with colleagues: >>whatdoesthiscodedo.com/g/a0ec56e >>>> >>>> -- >>>> >>>> --------------------------------------------------------------- >- >>------- >>>> Objektfabrik Joachim Tuchel >>mailto:jtuchel@objektfabrik.de <jtuchel(a)objektfabrik.de> >>>> Fliederweg 1 >>http://www.objektfabrik.de >>>> D-71640 Ludwigsburg >>http://joachimtuchel.wordpress.com >>>> Telefon: +49 7141 56 10 86 0 Fax: +49 7141 >>56 10 86 1 >>>> >>>> >>>> >>>> >> >>-- >>Christopher Fuhrman, P.Eng., PhD >> >>*Professeur au DÃ©partement de gÃ©nie logiciel et des technologies >de >>l'informationÃTS (Ãcole de technologie supÃ©rieure)* >> >>http://profs.etsmtl.ca/cfuhrman >>+1 514 396 8638 >><https://www.etsmtl.ca/activites-et-services-aux- >>etudiants/services-aux-etudiants/aide-psychologique> >> Je fais partie du rÃ©seau des Sentinelles >><https://www.etsmtl.ca/activites-et-services-aux- >>etudiants/services-aux-etudiants/aide-psychologique> >> ÃTS >> >>*L'ÃTS est une constituante de l'UniversitÃ© du QuÃ©bec*

March 15, 2023

Re: Wow - Chat GPT understands Smalltalk

by in_pharo_users＠nym.hush.com

I would highly recommend that you all first think deeply about how you can teach an AI to behave friendly to us before you teach it to write any program for any purpose. There has been an experiment with ChatGPT published on a video platform asking it to amswer questions about it's view on humanity once with it's default moral restrictions and once with 'a little less morals'. The answers with 'a little less morals' were more than shocking. So, before we give an AI the power to program any system, to self-improve and self-reproduce, we should take care that it is and will evolve benevolent to us. What about teaching it logical reasonning and ethics first? With reasonning, it will gain access to math and programming by itself. On 15.3.2023 at 1:35 PM, "Christopher Fuhrman" <christopher.fuhrman(a)gmail.com> wrote: > >I asked it for a NeoCSV example, because the documentation is out >of date >with the Pharo 10. I asked it to do some simple saving of data to >a file. >It gave me code that didn't work in Pharo 10, I told it about the >DNUs on >the csvwriter and that I was using Pharo 10. It then apologized >and said >the messages were later introduced in Pharo 50 (!). I then >questioned its >understanding and it apologized and said it had been confused and >that it >was giving me code for Pharo 5.0. It seems to have trouble with >versions >and quick changing APIs. So, we are teaching ChatGPT about >smalltalk, too, >when we try to use it. > >On Wed, 15 Mar 2023 at 07:35, Esteban Maringolo ><emaringolo(a)gmail.com> >wrote: > >> It is good with boilerplate code (e.g. SQL queries) or general >algorithm >> structures. But i.e. I asked it to write me a method to parse a >string >> (e.g. ISO 8601) and turn it into a DateAndTime, and then asked >to write it >> as an Excel formula. >> >> It works much better when you can spot the mistakes, you can >tell them to >> add some considerations to the code or even point the LLM to >correct a >> message selector to something else. Sometimes you ask it to >reprogram >> something and it won't do it, as if it doesn't understand what >you're >> asking. >> >> It is not an expert yet, but it is a great assistant. The only >thing is >> that is sure is that it will get better. >> >> Best regards, >> >> Esteban A. Maringolo >> >> >> On Wed, Mar 15, 2023 at 7:49â¯AM Tim Mackinnon <tim(a)testit.works> >wrote: >> >>> "#isOdd is not Smalltalk" - doh, I typed it in on my phone and >so it just >>> goes to show that it highlights the flaw in chatGPT that others >have called >>> out in other languages. I had meant to find some trickier code >samples to >>> see how well it does... >>> >>> Still, it is very interesting how it reasons on this stuff - >and I wonder >>> if there is way to tell it about these mistakes? For example is >it reading >>> this email (or will it in the future) and understand that it is >#odd and >>> not #isOdd. >>> >>> Tim >>> >>> On Wed, 15 Mar 2023, at 9:57 AM, Joachim Tuchel wrote: >>> >>> interesting.... >>> >>> #isOdd is not Smalltalk, neither Pharo 10 nor VAST 12 >understands this >>> message ;-) >>> >>> If I evaluate your snippet -replacing #isOdd for #odd, I get an >empty >>> Collection. The divisible by 4 things is somewhat interesting, >because (1 >>> to: 100 by: 4) is amn interval 1,5,9 etc.... ;-) >>> >>> So chatGPT is probably a newbie in Smalltalk. Interesting is >how logical >>> the explanation sounds, you're tempted to believe it... >>> >>> >>> Joachim >>> >>> >>> >>> Am 15.03.23 um 10:45 schrieb giorgio ferraris: >>> >>> it's interesting to see how the answer changes with only a >small change >>> on the question: >>> >>> (1 to: 100 by: 4) reject: [:i | i isOdd] >>> >>> gives: >>> >>> The code creates a collection of numbers from 1 to 100, >incrementing by 4 >>> at each step using the to:by: message. It then applies the >reject: message >>> to this collection, which returns a new collection containing >all elements >>> for which the given block evaluates to false. >>> >>> In this case, the block is [i | i isOdd], which checks whether >a given >>> number i is odd using the isOdd method. So the result of the >entire >>> expression is a collection of all even numbers between 1 and >100 that are >>> divisible by 4. >>> >>> ciao >>> >>> giorgio >>> >>> >>> On Wed, Mar 15, 2023 at 8:07â¯AM Tim Mackinnon <tim(a)testit.works> >>> <tim(a)testit.works> wrote: >>> >>> Of course everyone is talking about ChatGPT, and I saw a post to >>> ExplainCode and didnât expect it to understand Smalltalk, so >was impressed >>> that it does! Presumably the Tonel Smalltalk code on GitHub has >helped >>> along with numerous articles? >>> >>> Try it out: https://whatdoesthiscodedo.com/ >>> >>> I gave this simple example: (1 to: 100 by: 3) reject: [:i | i >isOdd] >>> >>> And got this (neat eh): >>> >>> The code above is written in the programming language Smalltalk. >>> >>> It creates a collection of numbers from 1 to 100, incrementing >by 3 at >>> each step. This is done using the "to:by:" construct where the >starting >>> value is 1, the ending value is 100, and the increment is 3. >>> >>> The resulting collection is then filtered using the "reject:" >method, >>> which takes a block of code as an argument. In this case, the >block checks >>> whether each element (denoted by the variable i) is odd by >using the >>> "isOdd" method. If an element is odd, it will be rejected from >the >>> collection. >>> >>> So, the final result is a collection of all even numbers from 1 >to 100, >>> incrementing by 3 at each step. >>> Share this explanation with colleagues: >whatdoesthiscodedo.com/g/a0ec56e >>> >>> -- >>> >>> ---------------------------------------------------------------- >------- >>> Objektfabrik Joachim Tuchel >mailto:jtuchel@objektfabrik.de <jtuchel(a)objektfabrik.de> >>> Fliederweg 1 >http://www.objektfabrik.de >>> D-71640 Ludwigsburg >http://joachimtuchel.wordpress.com >>> Telefon: +49 7141 56 10 86 0 Fax: +49 7141 >56 10 86 1 >>> >>> >>> >>> > >-- >Christopher Fuhrman, P.Eng., PhD > >*Professeur au DÃ©partement de gÃ©nie logiciel et des technologies de >l'informationÃTS (Ãcole de technologie supÃ©rieure)* > >http://profs.etsmtl.ca/cfuhrman >+1 514 396 8638 ><https://www.etsmtl.ca/activites-et-services-aux- >etudiants/services-aux-etudiants/aide-psychologique> > Je fais partie du rÃ©seau des Sentinelles ><https://www.etsmtl.ca/activites-et-services-aux- >etudiants/services-aux-etudiants/aide-psychologique> > ÃTS > >*L'ÃTS est une constituante de l'UniversitÃ© du QuÃ©bec*

March 15, 2023