If we ever build an AI that can think and feel, does Optimus’s logic apply to it automatically? Or is Megatron secretly onto something, that freedom has to be earned rather than assumed? And if an LLM ever crossed that line into real sentience, would we even be able to tell?
An LLM came up with a novel solution to the Erdős unit distance problem. It was unsolved for 80 years.
"And, on May 20, OpenAI shared a solution to the unit distance problem, along with a blog post explaining the work and a companion paper that featured nine world-class mathematicians commenting on the correctness of the proof and the importance of what had been done (as well as presenting a streamlined human version of the result). Mathematicians had generally believed that Erdős’ conjecture — about how many evenly spaced points can be placed on a plane — was correct. To general surprise, OpenAI’s internal model found a counterexample. To do so, it had found a sophisticated way to use tools from an area of math called algebraic number theory. As Jacob Tsimerman of the University of Toronto wrote in the companion article, “This is a really impressive piece of work. … It is definitely an intimidating construction.”
I can’t believe I have to keep explaining this to people, lmao
First of all, this is a math equation. Math is what computers do, they are essentially really fucking good calculators.
That math problem hadn’t been solved before, that’s true, however, if you actually read the excerpt you just shared, you’ll see that the equation was solved using preexisting theories and formulas. Ergo, humans had already manufactured every individual piece of the puzzle, and had used them in other contexts that the AI could be fed and then reuse.
Saying that AI came up with the solution isn’t the full story. That’s like saying that AI created a shitty piece of slop art. It did no such thing, it plagiarized a bunch of work from other people and essentially made it into a collage.
When MC Hammer (rather, his producer) sampled Rick James for “Can’t touch this”, did he become the bass player? No, that was still Rick James, they just took his song and altered it to make theirs.
That’s exactly how humans solve problems too. We don’t need to create entirely new languages and numeral systems, then write completely novel proofs for all the most basic theorems of math, before we can get credit for a new application of existing theorems that we discovered.
And yet people to this day know it as the riff from Can’t Touch This…
Hey dude, thought you might find this interesting: https://mastodon.social/@tristanbuckmaster/117233413705701198
Basically, remember that math equation that we were talking about AI ‘solving’?
I said it used pre-existing formulas rather than coming up with a novel solution, and you said that’s how all math works?
Well it turns out we were both wrong: A human solved the equation and published his work a few days before AI “solved” the equation.
So it didn’t even use pre-existing formulas, it just wholesale stole the answer from someone with actual intelligence, lmao
I’ll clarify that it wasn’t my intent to defend the AI hype. I agree that LLMs are just predictive algorithms that use probabilistic and non-deterministic heuristics to extrapolate likely responses.
I was just pointing out that saying they can’t come up with anything original is maybe not the best angle because it opens up a can of worms about whether even humans can come up with anything truly original, which is philosophically fraught.
Yes the mathematician solved the equation. But he couldn’t have done so without spending years devoted to learning the state of mathematics today, including all the work and research and theorems and proofs from all the mathematicians who came before him.
Maybe he applied old concepts in new ways to solve a problem that hadn’t been solved before. I doubt he invented an entirely new form of mathematics not based in anything that existed before. And even if he did that, chances are he only knew how to do that because of what he’s learned from pre-existing systems of mathematics.
That’s the point I was trying to make
Fair enough, I appreciate the clarification.
I would posit philosophically that the LLM cannot build off of the knowledge of others in the way that a human (like the mathematician) could, and so it is not capable of forming an original thought.
When an LLM model is trained, it is left in a sort of static state of “knowledge”. The model does not have “compounding memory” the way humans do; When a human finishes their education (training) and begins their career they usually continue to learn and gain experience throughout the course of their career.
An LLM doesn’t remember anything, not its experiences, not events that you tell it about, not the last time you prompted it. If you want to teach the LLM something new, you have to re-train the thing. It can’t learn on its own