Sylvia Else <
sylvia@email.invalid> wrote:
I've pretty much hit a wall with this experiment. Even within the same
session, getting ChatGPT to recognise that it's made a mistake does not
mean it won't make the same mistake again.
It's like trying to teach a dumb student something that is beyond them.
Even when you think they've finally got it, it turns out that they haven't. >>
And this is just with easy stuff. I have no hope that it would ever
learn to apply more complicated manipulations correctly.
Perhaps my whole approach is misconceived.
On further research[*] I think that last comment is correct. One is not actually teaching it anything during one of these sessions. One is
merely adding to the text that it will use as input to its neural
network to determine the next word to output. I wondered why its outputs
come as a slowish sequence of words, separated in time by significant intervals. I believe this is because during those intervals it is
determining the next most probable word to follow the previous words in
the session (both the user's inputs and AI's previous output).
So it can sometimes appear to be following instructions, but it's not
really doing that, and the more complicated the instruction, the less
likely the answer is to be correct.
This article suggests that in theory your principle of teaching
these AIs a new task via prompts is valid. It's called "in-context
learning". However as I understand it you need to teach the AI by
example rather than with explanations. The teaching process is
probably still a long way from being as easy as you were hoping
for, but theoretically possible in the right circumstances, and
apparantly sometimes easier than training a dedicated neural
network from scratch.
Solving a machine-learning mystery
by Adam Zewe, February 7, 2023
-
https://news.mit.edu/2023/large-language-models-in-context-learning-0207 "Large language models like OpenAI's GPT-3 are massive neural
networks that can generate human-like text, from poetry to
programming code. Trained using troves of internet data, these
machine-learning models take a small bit of input text and then
predict the text that is likely to come next.
But that's not all these models can do. Researchers are exploring a
curious phenomenon known as in-context learning, in which a large
language model learns to accomplish a task after seeing only a few
examples -- despite the fact that it wasn't trained for that task.
For instance, someone could feed the model several example
sentences and their sentiments (positive or negative), then prompt
it with a new sentence, and the model can give the correct
sentiment.
Typically, a machine-learning model like GPT-3 would need to be
retrained with new data for this new task. During this training
process, the model updates its parameters as it processes new
information to learn the task. But with in-context learning, the
model's parameters aren't updated, so it seems like the model
learns a new task without learning anything at all.
Scientists from MIT, Google Research, and Stanford University are
striving to unravel this mystery. They studied models that are very
similar to large language models to see how they can learn without
updating parameters.
The researchers' theoretical results show that these massive neural
network models are capable of containing smaller, simpler linear
models buried inside them. The large model could then implement a
simple learning algorithm to train this smaller, linear model to
complete a new task, using only information already contained
within the larger model. Its parameters remain fixed." ...
Research paper (not light reading):
https://arxiv.org/pdf/2211.15661.pdf
--
__ __
#_ < |\| |< _#
--- SoupGate-Win32 v1.05
* Origin: fsxNet Usenet Gateway (21:1/5)