shiandow

In a way I'm not entirely sure if proving the conjecture or posing it is the most important part here. It used to not matter much because proving results dependent on a connecture and making progress towards solving it were considered mostly equivalent.

But the distinction is going to become relevant very soon if many conjectures can be resolved (albeit in inscrutable fashion) by throwing raw computational resources at it.

I think they might be, some of them at least.

The notion that mathematics would be over just because it's slightly easier to find proofs is like the idea that astronomy would be over because of telescopes.

It does apply but they were allowed to have this data and GDPR does very little to protect it in that case.

While this discussion is fascinating I think we should scale it back a little. Instead of discussing all aspects I'm quite curious about just one.

How would you define "document" in a way that allows for formal verification? Ideally without relying on words that leave room for interpretation.

Code is a liability, generating more of it is solving the wrong problem.

Formalisation can't save you from determining what is and isn't a document. The judges main task is formalising reality and lawd, the rest of the inference is typically easy.

Is the museum supposed to have such a monopoly? Because then the ruling would be disappointing but ultimately not incoherent.

What's the law like on photographs? Those are effectively equivalent are they not? Or does the entire argument hinge on the fact that point clouds can be stored in a human readable format?

Honestly that makes me more convinced it's actually doing mathematics.

You can't outsource understanding, and to the extent LLMs can understand anything they can only learn through training. So until LLMs can truly learn, rather than bringing themselves up to speed with huge contexts, you'll probably still want some people who understand what it's doing.

I think Navier Stokes is a brilliant case study for all that it didn't do. An AI 'solved' a famous problem, of a kind that is impossible to cheat, yet the reaction is incredibly apathetic. Why? Because it apparently failed to do the thing people were hoping it would, but what exactly? Of course it also manages to fumble the delivery by antagonising the people most interested in the result, but still.

I guess I see how you could read it like that, but no I'm pretty sure this was meant literally.

> unless you mean a literal sticker warning when you buy meat from stores?

What else could it possibly mean?

You can't just expect things to last years! Building stuff to last only benefits society and customers, it does nothing for shareholders!

Experience is not understanding, but an LLM does not reason so it cannot understand.

It can produce text that looks like reasoning. It can even produce text with mostly sound logic, but there is no internal experience or reasoning that occured there just the generation of language.

LLMs therefore tend to be very bad at tasks that involve meta cognition. I've yet to successfully convince one to tell me when it knows something.

I understand less than before I read this article.

When has anyone ever noticed a full moon becoming less full during the night? Not that the effect doesn't happen, but it is minor. This effect also occurs during first and last quarter, despite what the article claims.

And what on earth is the first plot? A moon that rises to its zenith doesn't become half lit. And is the dark coloured part li?, Where is the 3/4 moon?

Huh, another reason to eat more spinach I suppose.

There comes a point where someone isn't so much talking to the sender as having an unsolicited AI chat. Especially when most of the information didn't come from the sender in the first place.

It's actually hard to define the difference between information that comes from the model and just random variation, maybe something to do with the cross entropy between the sender and the model?

Right but where is this information coming from, it can't come from the sender because an idea that can be conveyed in 300 bits cannot contain 1000 bits of information. You're just using the LLM to translate the idea into something more legible.

Alternatively those 700 bits are information the LLM added, but where is that information coming from? Is it noise? Random facts? Random lies? And who is the receiver even talking with if most of what they read is something the sender didn't know?

I'm not sure we're working in the same framing here. For one LLMs are random, sure you can fix the seed but you don't have to, you can even replace the RNG with true random noise.

So there are 2^300 possible ideas, only 2 outcomes from the cursory check, how do you get 2^700 outcomes? Most of those are just random variations the LLM added which is not a transfer of information. You would be lucky to even identify which of the 2^300 ideas was being conferred.

This is about information from the sender to the recipient. It is fundamentally impossible to transfer more than one bit in one binary decision.

A LLM adds noise, not information. At least in this framing.

I assumed a coin flip, indeed in practice you would likely achieve far less than 1 bit per attempt.

Unless that takes 700 attempts on average I don't think that actually works.

Unless that 700 bit was transferred on a separate occasion the inferred 700 bits is not true information, anyone could have reconstructed it from the 300 bits.

Well maybe not that, it would have to stop almost aligning text to make it good instead of usable.

I think it’s pretty unreasonable to use the service.

In what way are those two different? What makes the training data valuable if not to extract valuable information from it?

They sure as hell don't need it just to produce English.

If he didn't opt out I'm not sure I'd agree that it was fair game.

I'm pretty sure it would be considered plagiary amongst colleagues and it is a terrible precedent if we just let OpenAI steal any good idea they can get their hands on if they think it is profitable. You'd effectively sign away any and all rights to anything built with AI if OpenAI chooses to reengineer it before you.

And in all reasonable circumstances the court would agree. Poisoning the water supply is an act of war let alone an illegal act of violence.

The most interesting part to me is the bike. I don't think any of them would actually work, but the mistakes feel somewhat human.

Some of its standards supersede national law (I think it was the regulations that did so, amd standards needed to be built into national law, could be the other way around).

And this makes a lot of sense with rules about food safety, product standards etc. It almost makes sense for privacy protections. It makes very little sense for surveillance laws.