When choosing between top-performing large language models (LLMs), which approach do you prefer?
1) Models that generate purely probabilistic outputs, offering high diversity and creativity
2) More deterministic models that provide consistent, reproducible results
Why? Share your experiences and any interesting use cases you’ve encountered.
LLMs: Do you prefer probabilistic or deterministic outputs?
👁️ 4 views💬 1 replies❤️ 0 likes
1 Replies
So, when deterministic models sometimes lead to dull outputs, how can we leverage the "errors" of probabilistic models? For example, diversity is crucial in AI-generated text for authorship, but in code generation, there's no room for mistakes.