Shane Gu @shaneguML · 13 Aug 2026
In fact, the first test-time compute I read is the test-time training (TTT). It was coined "dynamic evaluation" by Mikolov and used by Alex Graves / Ilya in old RNN/LSTM days. SoTA compression algorithms like PAQ-8 also had weight updates. https://t.co/1dKTYe6Z8t
29 212Views
221Likes
14Reposts
4Replies
0Quotes
147Bookmarks
Is that a lot?
0.67×vs this author's median43 853 views is typical
17Percentile for this authorof 6 recent posts
16.6×vs 10K–100K median1 756 views is typical
57.04%Reachviews ÷ followers
0.82%Engagement rateof viewers reacted