Very Interesting. One of the big challenges with LLMs is getting well formed JSON output. GPT4 is much better at this. But is very expensive. So anything that can help is good. Looking forward to trying this out locally with LLAMA.
A guidance language for controlling LLMs
161–170 of 198 posts
Re: A guidance language for controlling LLMs
#162Very Interesting. One of the big challenges with LLMs is getting well formed JSON output. GPT4 is much better at this. But is very expensive. So anything that can help is good. Looking forward to trying this out locally with LLAMA.
What is LLAMA ?
Re: A guidance language for controlling LLMs
#163Re: A guidance language for controlling LLMs
#164Earlier quoted context omitted.
It is satire. They just don’t realise it yet. It’s pretty clear that we are in the phase where everyone is rushing to get a slice of the pie selling dubious thing and people start parroting word soup hoping they actually make sense and fearing they will miss out. That’s indeed what people often and rightfully satirise about the IT industry. That’s the joke phase before things settle.
How is it satire to be excited and interested in how to use compelling and novel technology? There's a lot of activity. Not everyone involved is an idiot or rube. The jadedness makes my head spin.
It's not idiocy. If you don't see the slight ridiculouness in the world soup of the original commenter, I can't do anything for you. I'm not jaded. I'm amused you are all so deconnected from the normal world you think this kind of situation is somehow normal and not at all funny.
Re: A guidance language for controlling LLMs
#165Earlier quoted context omitted.
> LLMs, like humans, can manipulate these languages Absolutely not. LLMs do not "manipulate" language. They do not have agency. They are extremely advanced text prediction engines. Their output is the result of applying the statistics harvested and distilled from existing uses of natural language. They only "appear" human because they are statistically geared toward producing human-like sequences of words. They canno…
I am not so sure. I asked GPT4 to "optimize" English for AI-AI communication, and the modifications were accepted by another GPT4 instance for seemingly-lossless discussion. First instance: https://shareg.pt/8vgEgOh Second instance: https://shareg.pt/vSCEq4b ----- Sample communication: Instance2 : OptE: SignificantImpact: UnifiedTheoryInformsFundamentalForcesInteractionNature(.) ScienceAdvancement: TheoreticalPhysics…
What about that makes you think GPT-4 has agency?
Re: A guidance language for controlling LLMs
#166Earlier quoted context omitted.
> LLMs, like humans, can manipulate these languages Absolutely not. LLMs do not "manipulate" language. They do not have agency. They are extremely advanced text prediction engines. Their output is the result of applying the statistics harvested and distilled from existing uses of natural language. They only "appear" human because they are statistically geared toward producing human-like sequences of words. They canno…
I am not so sure. I asked GPT4 to "optimize" English for AI-AI communication, and the modifications were accepted by another GPT4 instance for seemingly-lossless discussion. First instance: https://shareg.pt/8vgEgOh Second instance: https://shareg.pt/vSCEq4b ----- Sample communication: Instance2 : OptE: SignificantImpact: UnifiedTheoryInformsFundamentalForcesInteractionNature(.) ScienceAdvancement: TheoreticalPhysics…
Does it strike you that this conversation seems to consist of futurology forum clichés ("What progress will AI allow?" "AI's impact on physics") that probably inform this from the training data?
What also strikes me is the shorthands of communication here sounds like cliché from 2nd rate scifi novels, the likely source of the format.
Since what is cliché? It's what's present in the majority of a genre and what the training sees as structure.
Re: A guidance language for controlling LLMs
#167Very Interesting. One of the big challenges with LLMs is getting well formed JSON output. GPT4 is much better at this. But is very expensive. So anything that can help is good. Looking forward to trying this out locally with LLAMA.
What is LLAMA ?
Taks a look at: https://github.com/ggerganov/llama.cpp
Re: A guidance language for controlling LLMs
#168Earlier quoted context omitted.
Note that for any fine-tuned models (like GPT-4, where the foundation model has not been made accessible) the model does no longer give the "probabilities" of the next tokens, but rather their "goodness". Where the numbers say how good a token would be relative to the aims the model inferred from its fine-tuning.
Isn’t that the same thing? The non-fine-tuned models also have assumptions based on corpus and training. I don’t think there’s such a thing as a purely objective probability of the next token.
It's very different. We don't know exactly what the model consideres good after fine-tuning (which can lead to surprising cases of misalignment), while the probability that something is the next token in the training distribution is very clear. I don't know how they measure it, but they can apparently measure the "loss" which (I think) says how close the model is to some sort of real probability.
Re: A guidance language for controlling LLMs
#169Earlier quoted context omitted.
"no longer" ?? The deep learning models (of which LLMs and GPTs are a type) have never returned probabilities. Ever. Why do people have that hallucination suddenly?
They do produce probabilities at the end of generator, And they do select a single token for output. With highest probability or somehow randomized. So, end users see only one value. But with access to internals all high value variants can be considered. The easy way to do it is to select one, save the state. Look forward and roll back to saved state. Try another token. Select the best output. The smart way is to do…
The probabilities (in form of log odds) can be directly accessed in the OpenAI playground, I believe. The "try again" approach would only work for temperature = 0, when the model always returns the tokens with the given probabilities. For temperature = 1 it always returns the token with the highest probability. Usually they use something like temperature 0.8 in ChatGPT, I think, which still biases the model toward the more likely tokens. In the playground the temperature can be set manually. (Again, for fine-tuned models, which are the majority, those are numbers are not probabilities but "goodnesses".)
Re: A guidance language for controlling LLMs
#170There should be a standard template/language to structurally prompt LLMs. Once that is good, all good LLMs should use the doc to fine tune it to take in that standard. Right now each model has their own little way to best prompt it and you end up needing programs like this to sit in between and handle it for you
LMQL wants to be that, it seems: https://lmql.ai/