Unverified50% confidenceFactExact time
Selecting few-shot examples based on semantic similarity and increasing the number of examples can raise the LLM's F1 score on the hard test set by nearly 10 points over the baseline configuration.
1
Sources
50%
Confidence
Long-term
Relevance
11/9/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn class-imbalanced scenarios, the F1 score better reflects the model's ability to recognize the minority class than accuracy does.75% similarUnverifiedAn F1-score between 0.73 and 0.77 indicates that LLMs still exhibit errors in causal judgment, with considerable room for improvement.68% similarVerifiedStrategies for mitigating LLM judge instability include majority voting over multiple samples, cross-validation using multiple judge models, designing more fine-grained scoring dimensions, and retaining a manual review step.67% similarUnverifiedA single throughput number is influenced by factors such as batch size, precision settings, context length, and whether end-to-end latency is measured; this result should be understood as an extreme optimization level under specific conditions rather than a general judgment of hardware superiority.66% similarUnverifiedFor correction (rather than rewriting) tasks, small sequence taggers outperform small generative LLMs in precision, size, and speed.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1149995API
curl https://kongchang.com/api/v1/knowledge/claims/1149995MCP
get_claim(id=1149995)