Opaque Recurrence

What is opaque recurrence?

The reasoning we're used to has the AI "write out" its thinking — step by step, on screen (that's chain-of-thought). Opaque recurrence goes the other way: instead of spelling out the middle steps, it loops a computation through the model many times internally, tucking the thinking into the hidden layers and only handing back a clean answer.

How is it different from chain-of-thought?

Chain-of-thought: reasoning out loud
Every step is written down. It's transparent and checkable — but slow and token-hungry.
Opaque recurrence: reasoning in hiding
The middle steps never become text; the model just loops the computation inside. Faster and cheaper — but you can't see how it got there.

What does "opaque" mean here?

It's not an insult — it means the process is invisible to the user. Like watching someone stare at a problem for a moment and announce the answer, with the scratch paper taken away. That makes interpretability and auditing harder, and it's exactly what alignment research worries about: how do we know the hidden thinking is sound?

What problem does it solve?

Explicit reasoning burns a lot of tokens — it's costly and slow. Opaque recurrence lets a model do deep reasoning with fewer tokens and in less time, a fresh trade-off between depth and efficiency.

Bottom line: opaque recurrence is AI doing its thinking internally, looping the computation in the dark to get deeper reasoning for less cost.