Unverified50% confidenceFactExact time
Greedy decoding leads to a lack of output diversity, easily falls into repetitive loops, and often performs worse than temperature sampling or beam search on open-ended generation tasks.
1
Sources
50%
Confidence
Long-term
Relevance
11/2/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn tasks lacking repetitive patterns, such as free-form creative writing and open-ended question answering, the hit rate is low, and the speedup benefit drops significantly, sometimes even incurring slight overhead due to verification costs.72% similarUnverifiedWhen the marginal cost of generation and execution approaches zero, what becomes scarce is no longer the speed of writing code, but the ability to turn vague requirements into acceptable tasks, and the judgment about the output.72% similarUnverifiedMode collapse refers to the generator repeatedly outputting samples that easily fool the discriminator, resulting in severely insufficient diversity in the generated results.69% similarUnverified提高温度增加多样性往往以牺牲输出质量为代价，且在API调用场景下用户通常无法直接干预模型底层解码参数69% similarVerifiedThe non-determinism of large language models stems from their autoregressive sampling process; even with temperature set to 0, differences in inference frameworks, batch sizes, and hardware floating-point operations can all cause output drift.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1125286API
curl https://kongchang.com/api/v1/knowledge/claims/1125286MCP
get_claim(id=1125286)