What does a good language model look like internally in geometry?
We find a simple but surprising signal:
👉 the more spread out its hidden representations are, the better it predicts
(even for semantically similar contexts)
ICLR 2026 https://t.co/Yso57fMXmk
Presenting now👇
Introducing GPT-5.5
A new class of intelligence for real work and powering agents, built to understand complex goals, use tools, check its work, and carry more tasks through to completion. It marks a new way of getting computer work done.
Now available in ChatGPT and Codex.