This interview lasted over an hour. If he was answering wrong they should have said something.

I think this is disingenuous comparison. When we read a book we can estimate the amount of data we're taking in based on the character count (each character being represented by some fixed amount of bits).

What you're suggesting on the other hand is something akin to counting the number of pixels on each page we look at. That's absurd overestimate of the amount of data a person reading is actually taking in.

I believe there is a point: we simulataneously ingest words, but also glyph shapes and learn acceptable variations between them (eg. serif vs non-serif, large x-height vs small, curlier or more elegant, playful letters...) — all of these contribute to our multi-faceted learning, but ultimately, we do seem to need less of the data to learn (how long it takes for us to learn to recognize letters vs OCR based on ML).

I really don't think your argument is very convincing. Some of your examples are downright ridiculous? "Jobs" Making a movie about one of the creation of some of one of the most famous companies on Earth is obviously interesting. That's not comparable at all.

> Making a movie about one of the creation of some of one of the most famous companies on Earth is obviously interesting.

That's not what "Jobs" is about. It's the setting of a character-driven dramatization of Steve Job's life and his personality. You do not grow a company like Apple merely by dominating rooms with your personality, dropping oneliners and speaking in absolutes. There's a lot of actual work involved, the details of which would make for a movie rather boring to most audiences.

"Jobs" is a movie about Apple as much as "Inglourious Basterds" is a movie about the US military in WW2.

If you actually watched Jobs you would know it has very little to do with how Apple was actually created in real life.

Of course a movie about Steve Jobs is inherently more marketable than one about Qian Xuesen. But one isn't inherently more boring than the other.

I think there's been natural but steady progress with since 2024 with the release of the o1 model, which showed impressive reasoning capabilities. But I think it's wrong to look at the magnitude of the accomplishments and assume that will be field independent. We don't know the range of problems reasoning techniques are useful for. What we see here is refinement of capabilities that have been noticeable for years.

Maybe you need to phrase it better. Like with a more specific direction of thinking.

I think he's talking about reasoning models.

I think your analogy is good but I don't believe modern LLMs use Lean or any lean-like structure in their proofs. At least recent open source ones like DeepSeek can do advanced math without it (maybe the most cutting edge ones are doing it I can't say).

They are most likely using them in training. I doubt their IMO team are show ponies

Not a physicist either but my understanding is that is that if you believe that we can discover all the laws of physics that explain how the world operates then it needs to have a solution.

Like we have formulas describing how gravity works. We can test these formulas by observing the motion of the planets and galaxies. Is this theory true? There's lots of evidence for it so it feels like it's gotta be pretty close to "the truth"

We also have formulas describing how elementary particles behave. These formulas have been tested to a very high degree of precision so it seems they've got to be close to the truth as well. But if you use both our formulas for gravitation and formulas for elementary particles you can derive a contradiction. So these two theories cannot simultaneously be true. There's got to be something wrong with them.

I suppose there's the possibility that at a certain point nature simply doesn't follow any laws and you can't possibly make sense of it.

Are you sure? I think "normies" would prefer to see and try on the clothes they buy.

People will order clothes they see on tiktok without ever having touched them. Having something where their users can basically say "order me that shirt" while they are tiking their tok or rolling their reels, and it works most of the time, is a company's wet dream.

Though, people "want" a lot of things that actually end up making them less happy. So responding to demand doesn't necessarily make it a good thing, but only time will tell.

I can give you the exact mathematical formula used to statistically optimize the output of a neural network from input examples. Can you do the same for the brain?

Not atm, but does it matter?

Is it similar, even if much simpler, of the sort of process that goes on there too? That's the important question.