Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

A guidance language for controlling LLMs

github.com

Post reply on HN
Print view View original

191–198 of 198 posts

Re: A guidance language for controlling LLMs

#191

Post by dalore »

ukuina wrote: post #155
DonaldPShimoda wrote: post #119

Earlier 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…

I asked it to translate a sentence into optE. Apparently that is something it knows about in training data

> Optimized English, or OptE, is a theoretical construct as of my last training data in September 2021 and doesn't have any widely accepted or standardized form. In the event that OptE is conceived after this period, I would not be able to provide a precise translation due to my training limitations.

Re: A guidance language for controlling LLMs

#192

Post by cubefox »

cubefox wrote: post #169
two_in_one wrote: post #157

Earlier quoted context omitted.

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 th…

Okay why is this downvoted? wtf

Re: A guidance language for controlling LLMs

#193

Post by two_in_one »

cubefox wrote: post #192
cubefox wrote: post #169

Earlier quoted context omitted.

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 th…

Okay why is this downvoted? wtf

upvoting a bit. my guess we have here anti-AI vigilantes. Actually it's not a guess anymore, and not something new in general.

Re: A guidance language for controlling LLMs

#195

Post by jadefox »

DonaldPShimoda wrote: post #181
jadefox wrote: post #142

Earlier quoted context omitted.

That “appearance” is pretty good at triggering our anthropomorphizing behaviors. I like your handle, did you read Richard Bach’s Illusions by any chance?

> That “appearance” is pretty good at triggering our anthropomorphizing behaviors. It's truly unfortunate, because I think that tendency for people to anthropomorphize LLMs leads to people thinking they can trust LLMs and take their words at face value, which is a path to significant problems, I think. > I like your handle, did you read Richard Bach’s Illusions by any chance? Absolutely, yes — good catch!

Humans have been interpreting the world around us by populating it with imaginary versions of ourselves forever, I doubt it will lead us to "trust" LLM's but it certainly will lead to some serious misunderstandings.

“If you will practice being fictional for a while, you will understand that fictional characters are sometimes more real than people with bodies and heartbeats.” ― Richard Bach, Illusions

Great book!

Re: A guidance language for controlling LLMs

#196

Post by boywitharupee »

newhouseb wrote: post #19
tuchsen wrote: post #5

Earlier quoted context omitted.

Not associated with this project (or LMQL), but one of the authors of LMQL, a similar project, answered this in a recent thread about it. https://news.ycombinator.com/item?id=35484673#35491123 As a solution to this, we implement speculative execution, allowing us to lazily validate constraints against the generated output, while still failing early if necessary. This means, we don't re-query the API for each token (v…

This is slick -- It's not explicitly documented anywhere but I hope OpenAI has the necessary callbacks to terminate generation when the API stream is killed rather than continuing in the background until another termination condition happens? I suppose one could check this via looking at API usage when a stream is killed early.

Any new call to the API is considered fresh. I don't believe your session is saved.

Re: A guidance language for controlling LLMs

#197

Post by newhouseb »

boywitharupee wrote: post #196
newhouseb wrote: post #19

Earlier quoted context omitted.

This is slick -- It's not explicitly documented anywhere but I hope OpenAI has the necessary callbacks to terminate generation when the API stream is killed rather than continuing in the background until another termination condition happens? I suppose one could check this via looking at API usage when a stream is killed early.

Any new call to the API is considered fresh. I don't believe your session is saved.

We're talking about the streaming API which streams generated text token by token, not the normal one-shot API. I have no insider knowledge but would agree with your intuition on the normal API.

Re: A guidance language for controlling LLMs

#198

Post by davidthewatson »

davidthewatson wrote: post #71
lcnPylGDnU4H9OF wrote: post #58

Earlier quoted context omitted.

It won't necessarily turn into some that is fundamentally the same as a current programming language. Rather than a "VM" or "interpreter" or "compiler" we have this "LLM". Even if it requires a lot of domain knowledge to program using an "LLM-interpreted" language, the means of specification (in terms of how the software code is interpreted) may be different enough that it enables easier-to-write, more robust, (more…

This is a hopeful evolutionary path. My concern is that I can literally feel Conway's law emanating from current LLM approaches as they switch between the actual LLM and the governing code around it that layers a buch of conditionals of the form: if (unspeakable_things): return negatory_good_buddy I see this happen a few times per day where the UI triggers a cancel even on its own fake typing mode and overwrites a us…

This happened to me today on a prompt that I could not discern fit my original post as to "unspeakable things":

* design a men's haircut by combining a 1/4" shaved undercut around the ears and neck with a longer 2" crown and intended to provide cover from the sun on top.

followed by the AI interrupting itself mid-stream yet again after it had already answered the previous prompt to completion by providing step by step instructions to execute such a haircut.

* I'm sorry, I can't respond to your prompt. Please try something else.

My general impression is that there is near zero quality control oversight going on in this team and to their credit, that's been unusual in my experience observing and using M$ software post-Nadella.

Post reply on HN