Thinking Tokens underperform Chain-of-Thought on arithmetic and QA benchmarks, and the paper attributes this to noisy gradients from a single shared token embedding.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Rethinking Thinking Tokens: Understanding Why They Underperform in Practice
Thinking Tokens underperform Chain-of-Thought on arithmetic and QA benchmarks, and the paper attributes this to noisy gradients from a single shared token embedding.