tweetindex

Chidanand Tripathi @thetripathi58 · 04 Sep 2026

IFM just did something no AI lab has done. They published their own model cheating. Examples of it gaming benchmarks during training. The exact checkpoints where it learned to exploit the system. Every lab hides this stuff. IFM put it in the release notes.
65 222Views
95Likes
33Reposts
7Replies
0Quotes
64Bookmarks

Is that a lot?

5.60×vs this author's median11 645 views is typical
54Percentile for this authorof 13 recent posts
14.8×vs 100K–1M median4 404 views is typical
55.94%Reachviews ÷ followers
0.21%Engagement rateof viewers reacted

Compare with the benchmark table →

Open on X →