Unverified50% confidenceTradeoffExact time
Under the long-tail distribution of language tasks, the uncertainty-reduction gains from adding data may be far below what the 1/n intuition would suggest.
1
Sources
50%
Confidence
Long-term
Relevance
10/12/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAt equivalent parameter scales, diffusion language models typically still lag behind mature autoregressive models in perplexity and downstream task performance.76% similarUnverifiedEpistemic uncertainty stems from insufficient training data and can be reduced by adding more data, while aleatoric uncertainty stems from inherent noise in the data and cannot be eliminated by adding more data.76% similarUnverifiedCount-based prediction can partially identify which inputs are most sensitive to corpus changes, but substantially underestimates the rate at which divergence decreases in neural models.74% similarUnverifiedWhen information is insufficient, the content generated by large models tends to go off track; the poorer the context, the worse the output quality.74% similarUnverifiedIn information retrieval, increasing recall tends to decrease precision, and vice versa.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1001482API
curl https://kongchang.com/api/v1/knowledge/claims/1001482MCP
get_claim(id=1001482)