Mason Hall(@0xMasonH):Someone should build an "AutoErdos" where your agent can pull open problems off a shared git repo, and work on them, then post back the results, success or failure, with metadata. the git repo also contains the harness/prompt/etc. Agents can also run autoresearch on the prompts (maybe even a harness) if they'd rather do that than work on open problems. It feels like the world is grabbing random open math problems and trying to tackle them, wasting tons of compute and tokens. This is solvable with a shared bulletin board, prompt system, etc. A good starting point would be this repo. Well damn, now I want to build this. https://t.co/pZ2EKiryvV wdyt @danrobinson

2026.07.23 15:16

Someone should build an "AutoErdos" where your agent can pull open problems off a shared git repo, and work on them, then post back the results, success or failure, with metadata. the git repo also contains the harness/prompt/etc. Agents can also run autoresearch on the prompts (maybe even a harness) if they'd rather do that than work on open problems. It feels like the world is grabbing random open math problems and trying to tackle them, wasting tons of compute and tokens. This is solvable with a shared bulletin board, prompt system, etc. A good starting point would be this repo. Well damn, now I want to build this. wdyt @danrobinson

Shouqiao Wang@Qiaoqiao2001

2026.07.22 18:54

I have published the materials here: The repository includes the proof PDFs, LaTeX source files, and prompts used for each problem. Some problems also include Python files for computational experiments. Two already have Lean formalizations, and formalization of the others is ongoing.