Skullgrid comments on DO NOT talk about the goblins

[–] Skullgrid@lemmy.world 24 points 6 months ago (5 children)

And is the outcome good? Eh, sometimes.

this is the most damning fucking part of it. Oh, it's kind ok sometimes. Fucking hell.

It could be a shitload better, but that would be difficult to source accurate data instead of everything off github and stack overflow and let it fuckin rip bud. This fucking problem has existed since the LITERAL dawn of computing, garbage in, garbage out.

https://en.wikipedia.org/wiki/Garbage_in,_garbage_out#History

On two occasions I have been asked, "Pray, Mr. Babbage, if you put into the machine wrong figures, will the right answers come out?" ... I am not able rightly to apprehend the kind of confusion of ideas that could provoke such a question.

Pray tell, Mr Altman, if you were to feed the AI incorrect information, will the AI generate correct results?

permalink

fedilink

source

parent

hideshow 5 child comments

[–] snooggums@piefed.world 6 points 6 months ago* (4 children)

There is no magically reliable source of data that will make everything in one LLMs consistently accurate because their underlying design requires some randomization to reflect human conversation.

Dedicated models for specific use purposes where terminology is defined and they are designed to be deterministic would make them a lot better for actual use. We have had those models for years, just without the pretending to be conversational crap and they were constantly improving and actually useful.

permalink

fedilink

source

parent

hideshow 4 child comments

[–] wonderingwanderer@sopuli.xyz 1 point 6 months ago (3 children)

because their underlying design requires some randomization to reflect human conversation.

That's just false. Although the first step of creating an LLM from scratch is to generate a gaussian distribution, which is randomized, those matrices get overwritten multiple times throughout the process of pre-training and fine-tuning, when parametric weights are finely adjusted based on the training data.

During inferencing, tokens pass through various layers along specific embedded vectors weighted for relevance. It's not random at all. It's non-deterministic, but that's not the same thing as random.

If the training data all came from JSTOR or DevDocs or even WikiPedia, it's going to make much more accurate inferences than if it was trained on Reddit, Quora, and Yahoo Answers.

I'm not defending AI here, but lets keep our criticisms factual.

permalink

fedilink

source

parent

hideshow 3 child comments

[–] SparroHawc@piefed.world 2 points 6 months ago (2 children)

Except if you make the output token temperature too cold, it has a higher tendency to get stuck in loops and the like. A little bit of actual randomness is important.

permalink

fedilink

source

parent

hideshow 2 child comments

[–] wonderingwanderer@sopuli.xyz 1 point 6 months ago (1 child)

That's just adding noise, it's not unique to AI. It's also used in audio and visual design, and even cryptography.

permalink

fedilink

source

parent

hideshow 1 child comment

[–] SparroHawc@piefed.world 1 point 6 months ago

It's not unique to AI, no, but no one said it was. My point is that the noise is important to the functioning of the AI - and makes it even less deterministic, which also makes it poorly suited to automation in critical systems.

permalink

fedilink

source

parent