Looped Transformers Gain Attention as a Way to Scale Test-Time Compute

Aleksa Gordić argues that looped transformers, rumored to be used by frontier labs, are underrated: they scale test-time compute and reasoning without adding parameters or chain-of-thought, citing Geiping and Goldstein's work on latent recursion.

2026-09-17 ~ 2026-09-17 · 2 related posts

• Looped Transformer's Recurrent Depth: More Reasoning Without Extra Params or Longer CoT — gordic_aleksa · 2026-09-17
• Recurrent-depth latent reasoning scales test-time compute without extra params — gordic_aleksa · 2026-09-17