randomly doesn’t mean equiprobable. If you’re sampling a probability distribution, it’s random. Temperature 0 is never used, otherwise a lot of stuff would consistently hallucinate the exact same thing

if it’s allowed to use its own interactions as data, it will collapse. This has been studied. Stuff just does not work the way you think it does. Try coding one yourself.

The llm does not give you the next token. It gives you a probability distribution of what the next token coould be. Then, after the llm, that probability distribution is randomly sampled.

You could add billions of attention heads, it will still have an element of randomness in the end. Copilot or any other llm (past, present or future) do have this problem too. They all “hallucinate” (have a random element in choosing the next token)

turning jhonny into an llm does not work. because that’s not how the kid learns. kids don’t learn math by mimicking the answers. They learn math by learning the concept of numbers. What you just thought the llm is simply the answer to 2+2. Also, with llms there is no “next time” it’s a completely static model.

yeah. what’s your point. I said hallucinations are not a solvable problem with LLMs. You mentioned that alpaca used synthetic data successfully. By their own admissions, all the problems are still there. Some are worse.

from their own site:

Alpaca also exhibits several common deficiencies of language models, including hallucination, toxicity, and stereotypes. Hallucination in particular seems to be a common failure mode for Alpaca, even compared to text-davinci-003.

here’s that same conversation with a human:

“why is X?” “because y!” “you’re wrong” “then why the hell did you ask me for if you already know the answer?”

What you’re describing will train the network to get the wrong answer and then apologize better. It won’t train it to get the right answer

Yeah that implies that the other network(s) can tell right from wrong. Which they can’t. Because if they did the problem wouldn’t need solving.

the problem isn’t being pro ai. It’s people puling ai supposed ai capabilities out of their asses without having actually looked at a single line of code. This is obvious to anyone who has coded a neural network. Yes even to openai themselves, but if they let you believe that, then the money stops flowing. You simply can’t get an 8-ball to give the correct answer consistently. Because it’s fundamentally random.

no stigma has nothing to do with anything. people do die from its usage (or the consequences of its usage). It is more toxic than a lot of other “harder” drugs. It can ruin lives and break up families just as well. we’re talking about what it is, not how it is perceived. I am also not implying you should smoke weed instead. Everyone has their drug of choice. To each their own. But make no mistake, it’s a drug nonetheless.

yes it is, and it doesn’t work.

edit: too expand, if you’re generating data it’s an estimation. The network will learn the same biases and make the same mistakes and assumtlptions you did when enerating the data. Also, outliers won’t be in the set (because you didn’t know about them, so the network never sees any)

also, what you described has already been studied. Training an llm its own output completely destroys it, not makes it better.

The outputs of the nn are sampled using a random process. Probability distribution is decided by the llm, loaded die comes after the llm. No, it’s not solvable. Not with LLMs. not now, not ever.

no need for that subjective stuff. The objective explanation is very simple. The output of the llm is sampled using a random process. A loaded die with probabilities according to the llm’s output. It’s as simple as that. There is literally a random element that is both not part of the llm itself, yet required for its output to be of any use whatsoever.

why did it? because it’s intrinsic to how it works. This is not a solvable problem.

they really are not.

again, whoosh. you missed the part where you train me before asking the question. Then i can extrapolate. And I need very few examples, as little as 1.

I’m talking from the perspective of having actually coded this stuff, not just speculating. A neural network can interpolate, but it sure as hell can’t extrapolate anything that was not in its training.

Also, as a human, I can also train myself.

ok but that still entails trying random things until i find it. If I didn’t already know it was a builtin i wouldn’t know to search there. The bash thing was just an example. I have learned this stuff since i encountered the problem. This is just me recollecting my experience of trying to use man

I’ve met someone employed as a dev, who not only didn’t know that the compiler generates an executable file, but actually spent a month trying to change the code, not noticing that 0 of their code changes were having any effect whatsoever (because they kept running an old build of mine)

You point to me and tell me this is a bike. If we go around it 90 degrees and you ask me what it is, I can still tell you it’s a bike, even though I don’t know what one does or is used for. absolutely none of what you mentioned. i need no context. I only need to be able to tell that you pointed to the same object the second time even though I’m viewing it from a slightly different angle.

You point and say “this is a bike”, we walk around it, you point again and ask me “what is that?” I reply “a bike… you’ve just told me!”

Neural networks simply can’t do that. It won’t even recognize that it is the same object if it wasn’t specifically trained to recognze it from all angles. You’re talking about a completely different thing, which I never mentioned.