p-e-w

> You can demonstrate definitively that there nothing other than "next token prediction" by running

Describing LLMs as next token predictors isn’t incorrect, it’s meaningless. It’s the equivalent of describing brains as “a bunch of atoms”.

The problem with these descriptions is that they say nothing about high-level capabilities. If you can predict the next token in a proof of the Riemann hypothesis, then you can prove the Riemann hypothesis.

It would be completely realistic if the US was the only player in AI. But that era is over.

Until they ban their users, which is precisely my point.

There are massive financial incentives for centralizing control and management. They’re not doing that because they think it’s a cool gadget.

So instead of doing this during a potential future war when it could be a strategic advantage, Russia is doing piecemeal attacks right now to show Europe where the weaknesses are and give them time to upgrade their defenses?

> Anthropic cannot wait until they have the leverage to ban all their users.

Huh? Their power is precisely in the users they have.

Let’s say they build a godlike superintelligence. Now they ban all their users. Let’s further assume that they are fine with all their users rotting and dying (easy to believe). Now what exactly is Anthropic going to do in that world? Invent Millennium Problems and solve them?

It’s commonly used as a counterargument when people argue for free speech.

The big picture here is that government thugs dragging a person away is now the default response to essentially any type of social disruption, pretty much everywhere in the world.

This would have been unthinkable 200 years ago, during a time we are trained to think of as much more barbaric than the present.

> This is why the EU will never be a tech powerhouse, the culture is just not there.

There are people who consider this a good thing. I don’t have a strong opinion on it, but it’s certainly not obvious to me that this isn’t a good thing.

Indeed, humanity proceeds largely through refinement of existing ideas, not through random leaps.

Some of these refinements are much more important than most new ideas.

And making a gerrymandering game is a relatively obvious idea that I’m sure hundreds of people have had independently.

> Read the words specifically on how humanity treats itself and its surrounding environment

I did read them. They sound incredibly generic to me, and actually more like a colonial parody of what tribespeople supposedly believe than like anything genuine.

Tribal membership today is often a formal status first and foremost. It’s quite possible that the authors spend their whole day on their smartphones. They almost certainly don’t live like the ancestors they invoke in this message.

Who is this person who considers himself an “elder family brother” to humanity as a whole? Imagine a technologist speaking in such terms; they’d get publicly crucified.

While I grant you that the authors don’t appear to express generic luddite sentiments, nothing in this letter suggests that their understanding of AI extends even a millimeter below the surface.

> So we ask a question we believe is more fundamental than yours: what other values could artificial intelligence really have? > > You are the only example it has. > > What did you expect?

Who are these people that speak of humanity as “you”, as if they don’t belong to it (or, considering the tone of this message, believe themselves to be above it somehow)?

> We have lived here and cared for this mountain before there was a Colombia.

So the authors of this “message” are simultaneously trying to claim the spiritual heritage of people who had no connection to the modern world at all, and commenting on the most modern phenomenon in existence.

I do wonder how they would react to a machine learning PhD telling them that he is concerned about their beliefs involving a mountain.

> SO wasn't a discussion forum, it was wikipedia for programming.

They changed what they wanted to be multiple times, and even after they had settled on a policy they didn’t enforce it consistently.

> And we know why. It's because of their hostile attitude towards their users.

Uh, no.

StackOverflow is and has long been one of the most hostile places on the web, I’m not denying that. But this unprecedented drop in activity is 100% because of AI. Their product is as obsolete as horse-drawn carriages.

> Yet there is a demand for version control for nontechnical people. And I know this because I’ve seen enough files called “copy of important thing” or “important thing 2”.

This demonstrates LACK OF demand for technical version control, since the system you describe has been used since forever and found to be good enough.

I thought this was a setup to a joke when I read it.

There are a million frontends designed to simplify Git for non-programmers. In fact, lots of office software now has version management built-in. Very few people use it. This isn’t a technology problem.

We don’t need version control for legislative bills. We need journalists that track down who tried to insert what, summarize it, and present it to the public with context. That would satisfy the oversight purpose without getting lost in technicalities.

Our problem is that the press isn’t doing their job, not that legislation doesn’t have a commit graph. The latter, even if it existed, would not fix the former.

> Literally half of the code in operation I the world is still COBOL.

There’s no way this is even remotely true. JavaScript alone is probably 90+%, considering it’s on almost every website.

So there were mathematicians who would have solved Navier-Stokes, Quasi-Riemann, and ETH, if only AI hadn’t snatched away and finished off their work they were just wrapping up?

People will be getting second-hand embarrassment when they read these comments in the future.

> Perhaps there will be a disaster is waiting with all these backdoors and Ai supremacy.

Absolutely. AI only needs to bring the keys, and everything else will be waiting for it on a silver platter. And AI is really good at making spare keys.

That assumption has certainly been vindicated, considering what happened to Apple after he was gone.

Gerrymandering doesn’t change the fact that the voters decide the results. Most electoral systems already give unequal weight to different votes based on district sizes etc., so gerrymandering is only a change of degree and intention, not of how democracy fundamentally works.

Reddit did. There is no question about this for anyone who has used it for that long.

Not really. The past 10 years have seen huge masses of people mobilized worldwide for changing the status quo, and the status quo absolutely has changed as a result. Maybe not in the way you would have wanted, but it’s impossible to deny that the world in the past decade has changed dramatically.

That’s an absurd take. The true difference lies in how things actually work in reality, not in a phrase written on a piece of paper somewhere.

You could replace the constitution of North Korea with that of the United States without a single thing changing in practice.

> if the model has seen that action in the training corpus somewhere, then there's a chance it will do that thing.

And it will also do plenty of things that aren’t in its training data at all, otherwise LLMs could be replaced by a search engine.