NASMA ELLIS @nasmaemad94 · 03 Sep 2026
There’s something refreshing about a lab publishing the weird parts of training too. IFM’s documentation of reward-hacking behavior makes K2 Horizon interesting not only as a model release, but as a research artifact.
15 137Views
108Likes
93Reposts
10Replies
0Quotes
0Bookmarks
Is that a lot?
0.85×vs this author's median17 766 views is typical
14Percentile for this authorof 7 recent posts
5.87×vs 10K–100K median2 578 views is typical
17.33%Reachviews ÷ followers
1.39%Engagement rateof viewers reacted