Recent searches
Search options
An important thought that seems to elude the AI galaxy brains like billg who want to replace professionals with LLMs is: how are new phenomena identified and dealt with, given that LLMs are trained only on PRE-EXISTING data?
For example: how would a hypothetical LLM physician have identified a case of COVID19 back in February 2020? (Hint: it'd have diagnosed it as something else—wrongly—every time! Because not in the training data.)
@cstross That also means back-pressure on the tech industry, where "new phenomena" are people inventing things rather than observational science.
It may become impossible to launch a new programming language. No corpus of training data in the coding AI assistant; new developers don't want to use it because their assistant can't offer help; no critical mass of new users; language dies on the vine.
likewise the training pipeline to get people who are even capable of interpreting the alleged results - tho this is not a new problem; I myself have been fighting this for a decade
@zarfeblong @cstross I'm hearing: "If you want to hire quality programmers use an in house esolang for your business logic"
@seachaint @zarfeblong @cstross
You know, I’ve actually used “is writing a personal programming language” as a signal in a hiring decision.
@zarfeblong @cstross you think creation of training corpus is harder then convincing people to go and actually learn a new language?
@zarfeblong @cstross to flip that around, if you can supply and control the corpus for your language, you can ensure that people get better results with it than with other languages.
@zarfeblong @cstross that's a great shout. Holy crap.
#FBPE @zarfeblong @cstross But "writing chatbot prompts" *is* a "new programming language".
@TimWardCam @zarfeblong No, it's *playing at* being a new programming language. It's language design for toddlers, delivering the illusion of competence to MBAs who can barely write a buggy Excel spreadsheet.
@cstross @zarfeblong Y'know (and I was holding this back until I got some replies) that sounds very much like the design aim of COBOL. 
@TimWardCam @cstross @zarfeblong
No one in their right mind -- and I include the designers of COBOL here -- has ever inserted a mandatory random() in every major subroutine.
@dashdsrdash @cstross @zarfeblong Well, I've never used SPSS myself, but I've heard stories ... though those have in some cases been about the random element being in front of the keyboard.
@zarfeblong @cstross well there's a depressingly plausible thought.
@zarfeblong @cstross see, that one had never occurred to us because we had the differently-depressing thought that maybe it's the end of people learning things at all
@zarfeblong @danlyke @cstross Personally, this AI stuff has got me thinking about making a new language (or resurrecting an old one) for my own use. Who cares if it doesn’t get traction? I’m retired and unemployable due to bad attitude.
@zarfeblong @cstross Maybe a language without a corpus of examples to train AI on is exactly the language I want to use, because there is also not a corpus of AI generated slop in that language.