[flexagoon]

Is this not how current forms of reasoning work? It seems like the open models still output things like that, and the closed ones all just summarize their thinking instead to avoid distillation, but probably do the same thing internally.

[wrs]

I think the basic idea is the same (not being a frontier lab researcher I couldn’t say for sure), but there are different techniques, such as “reasoning tokens” that aren’t literally words, and more interesting structures than just sticking them into the stream.