New paper! Self-consistency is the unifying thread behind many exciting emerging training objectives: introspection, calibration, self-improvement, self-distillation, and more. We think it should be *the* organizing principle for the next generation of LM training.
New paper: It's time to optimize for 🔁 self-consistency 🔁
We’ve pushed LLMs to the limits of available data, yet failures like sycophancy and factual inconsistency persist.
We argue these stem from the same assumption: that behavior can be specified one I/O pair at a time. 🧵
Thank you @belindazli.bsky.social for the great talk "Solving the Specification Problem through Interaction” at our weekly seminar!
#NLProc
