4 ms·
1000 people using the same prompt will get 1000 different results. LLMs are completely non-deterministic.
by Kiro 4y ago
1000 people using the same prompt will get 1000 different results. LLMs are completely non-deterministic.
- richk449 4y agoWhy are they non deterministic? Is randomness intentionally injected? Or intrinsic to the approach?
- williamstein 4y agoCheck out the temperature docs in the API reference manual: https://platform.openai.com/docs/api-reference/completions/create#completions/create-temperature https://platform.openai.com/docs/api-reference/completions/c...
- alfiopuglisi 4y agoActually, LLMs are completely deterministic. Their output is a list of possible word, ordered by probability. If you always choose the highest-ranked word for the next iteration, they will always generate the same sequence for the same prompt. In all current implementations, a small amount of randomness (called "temperature") is added. In practice, the higher the temperature, the more probable is that a word down the list is selected, rather than the top one.
- dragontamer 4y agoI repeated the experiment 3 or 4 times before posting under slightly different phrases and sentences. I know this thing isn't deterministic. But never have I even gotten past step 2 of this blog post. I dunno. Has anyone else gotten the prompts to work as the blog post alleges? If so, can you share your prompts?