← Back to context

Comment by simonw

5 days ago

No, LLMs don't learn - each new conversation effectively clears the slate and resets them to their original state.

If you know what you're doing you can still "teach" them though, but it's on you to do that - you need to keep on iterating on things like the system prompt you are using and the context you feed in to the model.

6 comments

simonw

Reply
runarberg  5 days ago

This sounds like trying to glue on supervised learning post-hoc.

Makes me wonder if there had been equal investment into specialized tools which used more fine-tuned statistical methods (like supervised learning), that we would have something much better then LLMs.

I keep thinking about spell checkers and auto-translators, which have been using machine learning for a while, with pretty impressive results (unless I’m mistaken I think most of those use supervised learning models). I have no doubt we will start seeing companies replacing these proven models with an LLM and a noticeable reduction in quality.

rerdavies  5 days ago

That's mostly, but not completely true. There are various strategies to get LLMs to remember previous conversations. ChatGPT, for example, remembers (for some loose definition of "remembers") all previous conversations you've had with it.

  • runarberg  5 days ago

    I think if you use a very loose definition of learning: A stimuli which alters subsequent behavior you can claim this is learning. But if you tell a human to replace the word “is” with “are” in the next two sentences, this could hardly be considered learning, rather it is just following commands, even though it meets the previous loose definition. This is why in psychology we usually include some timescale for how long the altered behavior must last for it to be considered learning. A short-term altered behavior is usually called priming. But even then I wouldn’t even consider “following commands” to be neither priming nor learning, I would simply call it obeying.

    If an LLM learned something when you gave it commands, it would probably be reflected in some adjusted weights in some of its operational matrix. This is true of human learning, we strengthen some neural connection, and when we receive a similar stimuli in a similar situation sometime in the future, the new stimuli will follow a slightly different path along its neural pathway and result in a altered behavior (or at least have a greater probability of an altered behavior). For an LLM to “learn” I would like to see something similar.

    • rerdavies  4 days ago

      I think you have an overly strict definition of what "learning" means. ChatGPT now has memory that lasts beyond the lifetime of it's context buffer, and now has at least medium term memory. (Actually I'm not entirely sure that they are not just using long persistent context buffers, but anyway).

      Admittedly, you have to wrap LLMs to with stuff to get them to do that. If you want to rewrite the rules to excluded that then I will have to revise my statement that it is "mostly, but not completely true".

      :-P

      1 reply →