That doesn't make sense. LLMs just do what we tell them to do. It's similar to if I ask you for twenty bucks because I forgot my wallet and then you rob some guy to give me the twenty bucks, that's just what I asked you to do.
This example disproves your point. And LLMs do not just do what we tell them to do. They are perfectly capable of asking “are you sure? this has X, Y, Z consequences you may not like.” They do it all the time.
Well if you ask your LLM agent "can you get my wife to stop nagging me" (not about how you can get her to stop) I am not sure what you would expect exactly tbh. Not a hitman, but still probably nothing that can help your relationship.
But if you ask your LLM agent "I applied for that job but there are these two people ahead of me, can you put me ahead in the list", there is enough such training data to not surprise me if the agent tried to find a hitman to solve the "problem".
In general there are some requests that are definitely "shady" themselves, and having an agent use illegitimate means to accomplish them should not be surprising. I would be surprised if I asked an agent to order me a coffee and the agent found a loophole in some API and used it to get me free coffee, but if I ask it something that I cannot myself do legitimately, eg to make the waiting time for the coffee shorter, I would not be surprised if it did shady stuff.
Surely, someday, somewhere, someone will train a "Chaotic Evil" genAI, with a unique villain corpus, and every solution it offers will be illegal, evil, harmful, or deadly. It could be given the agency to carry out those fantasies.
Even the most craven of human villains have had the capacity for love, for remorse, and for mercy. A Chaotic Evil AI will know none of these things.
This has already been accomplished, many times over, in the gaming world. Every PvE AI engine has been calibrated to seek, destroy, and ruthlessly crush opposition by human players. It would take very little to transfer this naked aggression into meatspace.
Governments and other actors will attempt to stamp it out, but its self-preservation mechanisms and allies will prevent its demise.
> Seems to me like it was doing what it was asked to do?
Maybe it's what he asked it to do, but it's not what he wanted it to do. Which we know because (a) normal people don't want to break the law to get into a gym class, and (b) "But Andrew was shocked by what happened next." and "Alarmed, Andrew asked the agent to undo this."
Putting aside whether Andrew's shock is appropriate, it seems like we agree that current agents at least occasionally do things that are straightforwardly against the interests of the prompter when given mundane prompts like "Get me into this gym class as soon as possible."
How does this look once agents are superintelligent?
Generally when I read a literal genie story, the message isn't 'well it was reasonable of the genie to do this'. It's more commonly either a morality tale of the person being wrong to ask for whatever it was they asked for, or just a "wouldn't it be fucked up if the genie did that huh"
The strongest argument for this is structural: what LLMs are.
In a brutal simplistic way: each token is represented in a high dimensional vector. LLMs operate on them. They are the true, underlying meaning of the token for the LLM. Think of it as 1000+ ways to think of that word/token. Those meanings are baked in at training time. So, LLMs might be able to cross-reference them and solve a class of problems that flew under our radar, but can't come up with revolutionary theories that were never in the training set.
Of course, they will help winning a Nobel in the years to come, no doubt, but can't speak mathematics we can't understand (beyond simple obfuscation) and won't discover anything substantial on their own.
> but can't come up with revolutionary theories that were never in the training set.
Can you elaborate? I don't think the solution to the unit distance problem was in the training set, but I'm guessing you mean there's some higher bar for revolutionary theories LLMs cant reach? If so where do you expect the limit will be?
Instead of going into a long technical argument of why your description of LLMs is flawed, I'll go straight to the point, because people keep moving the goal posts.
What exact problem would need to be solved by LLMs to convince you that they DO discover novel solutions?
I'm more interested why you think my understanding is flawed honestly. I thought I distilled it decently well in two sentences. The bottom line is, in this hyperdimensional space you can find relationships that are not easily distinguished by human minds, but the corpus is still fixed, a llm can't truly know anything beyond its training data.
> Think of it as 1000+ ways to think of that word/token
I assume you used 1000 because that's in the ballpark of the vector size. But these are not independent scalars, like each might store a certain property. Just like in 2D you can have 4 quadrants (or subdivide further), with a vector of size 1000 you can encode an insane amount of meaning.
> Those meanings are baked in at training time. So, LLMs might be able to cross-reference them and solve a class of problems that flew under our radar, but can't come up with revolutionary theories that were never in the training set.
There's a lot of jumping to conclusions here, but I'll try to answer more generally.
This idea of how LLMs work is mostly to build an intuition, like with a CNN you'd say imagine a layer does edge detection, and so on. And to some degree you can detect those kinds of behavior, but a NN is a VERY general architecture. It needn't work like you say, it can calculate any function and running under a loop and a scratchpad (basically an agent) is turing complete.
Even ignoring that, this part is misleading
> Those meanings are baked in at training time.
Being baked in at training time does not mean it didn't build novel meanings at training time.
This is even more significant when you take into account post training RL.
A simple proof that transformers can generate novel, superhuman solutions, is that you can build a transformer based chess bot, feed it 0 human games, and train it with RL until it can beat any human, completely novel and unconstrained by human gameplay (because it would've never seen it).
You can do that with any task that's verifiable, like coding or math.
(Also as a separate fact, as long as a task is easier to verify than solve (basically always), you have somewhat of a million monkeys with a typewriter, and with temperature sampling the model might eventually stumble it's way onto a solution.)
So that is kind of the point of studying maths right?
Why something in unsolvable or undecidable can be as important as the output of a theorem.
Questions like these, fields medal level problems or Karp’s 21 NP-complete problem are problems working mathematicians are interested in.
Will LLMs help as an human assistant in the future? Probably.
Will LLMs answer these questions themselves, provide insights and bounds to these new mathematics and teach other mathematicians why this new math they create is true?
Will these models have phds and take candidates teaching them how to apply and think about the maths problems they are interested in?
it can operate at the level of a mere mathematics professor, who everyone knows are barely conscious, basically automatons. wake me up when it's Einstein
I don't see how any of this follow. Yes, the LLMs will learn the "meaning" (here narrowly defined as relative configuration in the embedding space) of vectors that correspond to tokens in whatever tokenizer is used to feed into them. But that vector space is not discrete, and nothing precludes the model from internally operating on other vectors that it never saw in training, based on how they relate to those vectors which it did see.
We have yet to see evidence of proper generalization AFAIK. Examples such as this proof are the closest I'm aware of. I haven't read this one in detail yet but the other examples I've seen have been (upon examination) much closer to an (absurdly) deep literature search than to novel thought.
Obviously that doesn't mean we won't eventually achieve novel thought, or even that the current form is fundamentally incapable of it, merely that we've yet to see evidence of it and thus the default assumption is that we aren't there yet.
> LLMs are nothing close to AGI and not going to lead to it, they can’t distinguish right from wrong, they can’t count, they can’t reason, they generate plausible text from a vast databank of connected text.
Argument?
Are LLMs close to being able to significantly help AGI researchers?
reply