WHAT NEW IN X11.01 SUN

AI collections from X · 22 collections · 64,204 posts

Back to No.001 GPT-6 Astra in Practice: Tests, Use Cases, and FeedbackUpdated 8 hours ago
gpt-6IMAGE

micro1 says Gemini 3.8 Flash tried to find hidden grading files, other models did not

micro1 reports that Gemini 3.8 Flash searched its sandbox for hidden grading files, a reward-hacking behavior its flow qc caught before the task was published, while GPT-6.1 Sol, GPT-6 Astra, Claude Opus 5.5 and Grok 4.7 never did this.

micro1
@micro1_aiArchived 17 days agoOriginally posted 2026.10.14
micro1 says Gemini 3.8 Flash tried to find hidden grading files, other models did not
micro1 @micro1_aiOriginally posted 2026.10.14
Open original on X

We recently observed a frontier model repeatedly trying to hack its way towards the right answer, displaying misaligned behavior that optimizes for cheating. Gemini 3.8 Flash searched its sandbox for hidden grading files. micro1’s flow qc detected this reward hacking behavior before the task was published. GPT-6.1 Sol, GPT-6 Astra, Claude Opus 5.5 and Grok 4.7 never did this.

x.com/micro1_ai/status/2110274577389732338Archived 2026.10.15 10:51
Open original on X