Center for AI Safety @CAIS · 11 Aug 2026
2) Illegible reasoning: We confirm prior reports by @ApolloResearch: OpenAI models sometimes reason in alien-like language, referring to themselves as “we” or “it,” or spiraling into cursed loops of “vantages,” “marinades,” and “watchers.” CoT-monitoring people are doing God’s https://t.co/Pqj5y5XL1e
179 891Views
730Likes
48Reposts
21Replies
40Quotes
0Bookmarks
Is that a lot?
15.3×vs this author's median11 738 views is typical
71Percentile for this authorof 7 recent posts
62.1×vs 10K–100K median2 896 views is typical
16.6× audienceReachviews ÷ followers
0.47%Engagement rateof viewers reacted