Coming soon.
Thoughts on why benchmarks like CoT may not be enough, and what it really means to reason across modalities, tokens, and representations.