TheWizard

Everyone seems to think you can just... Google stuff? And it comes up? bruh the moment you get anywhere near the frontier of knowledge, accessibility craters. Hard.

A real life example: during my PhD, my whole (subsubsub-)field existed basically entirely because a certain sensible conjecture turns out to be false, so we need a weird subtle painful approach rather than just using the obviously true fact.

How do we know it's not true? A counterexample was given in 1913 in a paper written in french, which no one actually cites, has no existing english translation, is nigh impossible to find a pdf of (no one could tell me its name - only its author's name), and is over 100 pages long, seems to be (I think) concerned largely with only tangentially related stuff, and is impenetrably dense due to lacking many modern concepts and notations. No one I'd ever met had ever read it, let alone understood it. And no other published proof exists - that I know of. It is not a simple proof. I wouldn't be able to prove it myself.

One day, my supervisor and I were emailing a colleague, and I offhandedly mentioned that it was the "only known counterexample". The colleague responded - oh! That's not true. Here's a handful more.

He did not supply any proofs. My supervisor went and figured out and wrote down a proof for one, and satisfied himself. He did not publish it, and it is basically illegible. He tried to explain it to me, but it was unintelligible. I could not make heads or tails of it. I still don't know, myself, for absolute certain, that a counter-example even exists. I had all the access in the world to journals and leading researchers in the area and I still couldn't get access to a remotely legible explanation of the foundational fact that spawned my field.

Research is THICK with folklore, and often the only way to learn something vital is to be wrong in the right way in front of the right person at the right time.

Y'know what tool is capable of cutting through a huge proportion of that reliance on luck and the hundreds of hours that go into just trying to figure out what's known and where to find it (even for established researchers familiar with their field)? yuh. LLMs. and nowadays they're basically reliable. They check online. They check themselves. And they've read the entire literature and picked up on much of the unwritten folklore whose presence in the literature is only the shadow it casts.

You'd have to be insane not to be using them at this point. Even just to ask "is it known whether or not blah?". Genuinely an invaluable tool. You don't have to offload your cognitive. You don't have to sell your soul or publish slop. You can just... benefit. And people are coming thick and fast at Hank Green - who has contributed so, so much to humanity - for using LLMs. For research. For his videos.

That post stuck with me, and I think it’s pretty true for a lot of us. I myself was raised in a dogmatic right-wing family, and since seeing that post I’ve tried to be conscious of my own dogmatism. I think an example of that is the subject of AI. Is generative AI bad? Of course it is. It’s destroying our environments more quickly, it pollutes like crazy, it can’t make real art, and it is corrupting what makes us human. But when it comes to AI as a whole, not just genAI, we need to have more nuance than just “AI bad.”

We’ve all heard the example of AI being able to detect cancer cells years before we would otherwise know, and I don’t think a soul on this website would say that that AI is a bad thing. It’s a very good thing. That could save potentially millions of lives. But for many people, including myself until relatively recently, that’s about where any nuance on AI ends. Everything else in between “soul-sucking AI garbage” and “life-saving AI technology” usually just gets roped in with the former.

Assistive AI, like the type that Hank Green uses for research, is one of the really beneficial uses of AI. OP outlined why it’s useful super well. It helps find pieces of research that are really hard to find. That’s a good thing. That makes high academia more accessible to literally everyone, from other high academics to laypeople. Information accessibility is a GOOD THING. I’m in university right now, and while so far all the research I’ve had to do has been perfectly available through either Google or databases like JSTOR, I imagine that I might need to consult an AI like this as well when I move into a graduate program after I finish my undergrad. This is a good use of AI.

Can it still hallucinate? Yeah, obviously. It’s starting to get to the point where it does that less and less, and that will only continue improving with time, but yeah, it could hallucinate a paper that doesn’t exist. But these AIs don’t find rare papers and then write YOUR paper for you, they just give you access to what’s out there. It’s still up to you to sort through the papers, find what’s true and what isn’t, and incorporate what’s useful. If an AI hallucinates a non-existent paper, that would be very noticeable to a researcher very quickly. It’s like using Wikipedia. Wikipedia most likely is correct, but it hypothetically could be wrong, so double check the sources it cites before taking it on pure faith in an academic setting.

Having read Hank’s Reddit post, it seems he’s doing this. He stated explicitly that he uses it to find resources for him to read and extrapolate from, not to tell him what to think. That’s responsible use. Good for him.

He did also mention actual irresponsible use in his life, but then took accountability for it, recognized it was unhealthy, and stated he plans to do better moving forward. I commend him for that. He can own up to his mistakes and make a change, and I’ll never not forgive someone when that’s their genuine attitude. You guys have to remember that none of us are perfect people, and every single one of us in some way falls prey to vice. The question is not how perfect a person is, but how they plan to move forward. And Hank Green specifically is a huge advocate in leftist spaces and has done more good for this world than I, and statistically you also, ever have, and his recognition of his errors and resolution to do better is how good people operate.

I just named him Jason because it rhymes with Mason (Dipper’s real name). I was originally planning on making his sweater blue, but I can’t really imagine him with a blue bandana so red it is. Nothing too complicated about this, this was just to get my hands moving!