Intuitively, regex or json grammar have a much lower "semantic dimension" than what today LLMs allow. Maybe the observed performance gains result from such lower dimensionality.
That whole structured generation line of work looks promising. I hope someone else takes this and runs evaluations on other benchmarks. Curious to see if the results translate!
Agreed! While these results are very promising, there's still a lot to explore in this space.
In addition to the "prompt consistency" and "thought-control" ideas mentioned in the post, I'm definitely curious how the performance is on more complex structured data (things like codegen).
