Remix.run Logo
▲ mahogany 5 hours ago

Probably closer to 90/10

▲ks2048 an hour ago | parent | next [-]

I'm not sure numbers like this have much meaning.

It's worth remembering that this was done by someone who (as far as I can tell) is a great mathematician and not by some "prompt engineer" with an interest in math and large budget.

▲suddenlybananas 5 hours ago | parent | prev [-]

None of us can know this without all the prompts/history

▲kingstnap 2 hours ago | parent | next [-]

TFA literally says this:

> but this time Claude far outpaced me near the finish line

> I intend to do my best to understand and impart understanding of this argument and its certain simplifications, to speak about it, to write it properly

> we were going to win but I was also, so to speak, out of my element, in that I personally have little mastery of automorphic forms on higher rank groups

> I’m just too inexpert in the relevant stuff at the moment, I hope the reader sees the breadth of this argument!

Also the exposition at the bottom of the paper is straight up just claude.

> So here is a rough timeline Claude provided

Also its like literally gigabytes worth of lean... it goes without saying that that is going to be effectively 100% claude.

I'm sure Levent put in some good work, but also this is clearly in large part the work of claude.

▲mahogany 3 hours ago | parent | prev [-]

Sure, and it depends on what we mean by the ratio. But, seeing that the proof comprises millions of lines of lean code (which have not been read or understood by a human), or 250k pages of prose (which have not been read or understood by a human), it makes me think the machine largely "did" the work of this proof. How far do you think he would have gotten without the machine?

Actually, thinking about it more, the ratio might be closer to 99/1.