| ▲ | seizethecheese 3 hours ago |
| Was this solved in a chat interface? It's not clear, but it seems like it from the document. The most relevant quote seems to be: > Claude and I have had a great time together with such breadth in recent months, but this time Claude far outpaced me near the finish line. No, Claude flew, and danced, and sparkled near that finish line. I had us on the line of combining two quite different ideas present in a toy setting, one of which was already new and shocking, making very simple one of the shadows of Riemann over which we
mathematicians felt most triumph. I guided us to the right situation analogous
to that toy setting, and then Claude began to leave me in awe. If indeed this was simple 1:1 chat, that's quite important, in and of itself, in my opinion. Presumably, OpenAI and others have been throwing autonomous agent swarms at this problem without success, but a single human + Claude seems to have cracked it. We may be in the "centaur chess" era of superhuman intelligence. |
|
| ▲ | tnrich 3 hours ago | parent | next [-] |
| just cause you're talking w one agent doesn't mean that agent isn't coordinating 100's (or 1000's) of sub agents to tackle pieces of the problem.. |
|
| ▲ | FrasiertheLion 3 hours ago | parent | prev | next [-] |
| What about that line implies chat? |
| |
| ▲ | seizethecheese 3 hours ago | parent [-] | | It sounds quite collaborative. I would assume that spinning an autonomous swarm would not be described this way. It's not clear, for sure, but it sounds like it to me. | | |
| ▲ | becquerel 3 hours ago | parent [-] | | I would imagine it's an agent swarm with a dedicated coordinator or human interface agent - the one agent you actually talk to who interfaces with the ten morbillion other ones. It's perhaps strange to treat that swarm as a gestalt entity called Claude, but hey, we're in strange times. | | |
| ▲ | seizethecheese 3 hours ago | parent [-] | | Yeah, if it's a swarm and they're referring to it as an intimate conversation with Claude, that would be interesting. | | |
| ▲ | mkipper 2 hours ago | parent [-] | | It’s possible the mathematicians at Anthropic don’t even know the details of how this works. If Anthropic is willing to hire a bunch of mathematicians to solve open problems, they’re probably also willing to have some engineers help set up the infrastructure for them to work effectively. So the researchers might see something resembling a regular chat and not realize that every non-trivial prompt related to the problem spins up a swarm of agents and burns $50k in tokens. That’s just speculation, but given the money involved in this stuff, it’s hard to imagine Anthropic is just sitting these mathematicians at a computer with a chat window and unlimited token budget and leaving them alone. They’ve probably sunk lots of engineering time into supporting this research. |
|
|
|
|
|
| ▲ | i_no_can_eat 3 hours ago | parent | prev | next [-] |
| Obviously not. Did you miss the part of the thousands of lines of Lean code? |
| |
| ▲ | seizethecheese 3 hours ago | parent [-] | | The linked document says the proof is incredibly long, but nothing else, so yes I guess I missed it. "Thousands of lines" doesn't exclude chat at all. You can easily generate this in a single chat session in an afternoon for coding. Do you mean millions? | | |
| ▲ | i_no_can_eat an hour ago | parent | next [-] | | I meant thousands of pages, millions of lines, yes. | |
| ▲ | bigmadshoe 2 hours ago | parent | prev | next [-] | | I read elsewhere in this thread it is 15M lines of lean. | |
| ▲ | GrollingGrakbor 2 hours ago | parent | prev [-] | | It's 17M according to the github repo. Which btw, is akin to showing someone the bible and saying "Proof of God is in here. It may be 2000 pages and in Latin, which you can't read, but just believe me." |
|
|
|
| ▲ | bigmadshoe 2 hours ago | parent | prev [-] |
| I don't think a single instance of Claude wrote 15M lines of lean code or produced a 250k+ page "slop proof" when prompted. |