Autoresearch user reports a 12-day run using 34 billion tokens
The user says intermittent failures, including race conditions and reward hacking, are a major issue because they may surface late and require significant rollbacks.
TLDR
A user reports 12 days of execution time and 34 billion tokens for their primary GPUMODE autoresearch campaign on the QR problem. They highlight intermittent failures, such as race conditions and reward hacking, that may only be discovered later—requiring significant rollbacks.
Combined views
57
1 Source, first seen 17d ago
reposts