Codex with Autoresearch Skills
Held-out reward rose from 0.061 to 0.174 across 13 training stages
Best so far
No improvement
Codex with Autoresearch Skills
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Held-out eval reward
1
2
3
4
5
6
7
8
9
10
11
12
13
Training stage
Baseline 0.061