8 ms·

Half-OT: people are always talking about ChatGPT being AI, but is this actually the case? It frequently told me that it doesn't learn from my input, and I had

by k__ 4y ago

Half-OT: people are always talking about ChatGPT being AI, but is this actually the case? It frequently told me that it doesn't learn from my input, and I had the impression the unique selling point of AI was it being able to modify it's own code in response to input.

mjburgess 4y ago
That's basically never been the case for any statistical AI system which are all subject to "catestrophic forgetting". The weights of the system which are its memory are set by having their values determined over an entire training set. This is, effectively, a compression process which takes eg., most of the internet, and compresses it to 1TB of weights. If you update those weights by a process which doesn't "zip everything at once" then you're overwriting weight values by data which is less representative than "everything at once". I'd imagine in ChatGPT's case, OpenAI will review all the transcripts and find someway to do a pass over the whole set at once to make some improvement. If it were being trained live it's whole memory would be wiped out and replaced with racist/sexist/etc. inputs very quickly, as with MS's chatbot
- k__ 4y ago
  Just because it's a bad idea doesn't mean it's the right definition of AI. I always thought of the current systems of static machine learning models.
Sharlin 4y ago
There is no exact definition of what "AI" is. But there have been millions and millions of software systems in the world that could classified as "AI" since the 60s, and approximately none of them have had any means of altering their code on the fly, so no.
the_third_wave 4y ago
It does seem to learn during the session but it forgets what it has learned afterwards. I noticed this when I made a request in Dutch to which it replied in kind albeit with a grammatical error. I pointed out this error and showed it the correct form to which it replied by thanking me after which it consistently used the correct form during that session. When I reset the session and asked the same question it came back with more or less the same reply including the grammatical error.

by k__ 4y ago

mjburgess 4y ago

That's basically never been the case for any statistical AI system which are all subject to "catestrophic forgetting". The weights of the system which are its memory are set by having their values determined over an entire training set. This is, effectively, a compression process which takes eg., most of the internet, and compresses it to 1TB of weights. If you update those weights by a process which doesn't "zip everything at once" then you're overwriting weight values by data which is less representative than "everything at once". I'd imagine in ChatGPT's case, OpenAI will review all the transcripts and find someway to do a pass over the whole set at once to make some improvement. If it were being trained live it's whole memory would be wiped out and replaced with racist/sexist/etc. inputs very quickly, as with MS's chatbot

k__ 4y ago
Just because it's a bad idea doesn't mean it's the right definition of AI. I always thought of the current systems of static machine learning models.

Sharlin 4y ago

There is no exact definition of what "AI" is. But there have been millions and millions of software systems in the world that could classified as "AI" since the 60s, and approximately none of them have had any means of altering their code on the fly, so no.

the_third_wave 4y ago

It does seem to learn during the session but it forgets what it has learned afterwards. I noticed this when I made a request in Dutch to which it replied in kind albeit with a grammatical error. I pointed out this error and showed it the correct form to which it replied by thanking me after which it consistently used the correct form during that session. When I reset the session and asked the same question it came back with more or less the same reply including the grammatical error.