6 ms·
What all these tools need to adopt is sending 10-20 requests out and finding the "best" response. I think it's incorrect that we try to get the tool to work rig
by wjessup 3y ago
What all these tools need to adopt is sending 10-20 requests out and finding the "best" response. I think it's incorrect that we try to get the tool to work right the first time. Auto-GPT has JSON parse errors 20-50% of the time. Instead, with enough parallel responses we can increase the likelihood one of them is "really good". The next challenge is figuring out which answer is really good and continuing with that.
- mpaepper 3y agoI thought about that as well. Some kind of ensemble agent which is more robust, might play with that idea.
- typon 3y agoEach with a different temperature too - kind of like a multi-armed bandit but exploring different parts of the LLM manifold.
- sharemywin 3y agoyou could probably get the agent to create "semantically similar but syntactically different prompts" to create more variation in the response.