Sharing insights regarding prompts

[fire17]

trafficstars

Hi there, since this is an evolving system, i think we should use collective experience, and like a tesla fleet, get better collectively

Proposal: Record Insights, Submit for review, Use Pinecone Embeddings

if a task is requested and it takes it X iterations to solve it, what if we can make a report of insights might save us steps in the future, should we try again, or advise others to replicate. There are 7k+ people who started and might be using Auto-GPT, together, the learning curve can be increadible One important thing to remember, is to check each submit and flag potentially harmful insite reports Then we need to come up with a rating mechanism, like stars, flags, something like stackoverflow, that agents can mark if some insight was helpful, or not helpful, etc, and this database of solutions/insights will be a fluid and evolving knowledge base might even be wise to allocate "repositories" or "namespaces" that can be verified by the comunity, the repo can be a client to a public pinecone server, that is managed and verified by the comunity and only accepts verified insight submissions

Like a good fleet, we need to share with eachother all the safe and useful progress, and flag bad practices

If you get my drift, please tag relevant people you think are up for it If you're worried about token costs, remember this might save multi steps, and also we should have gpt4all soon, and it can also just be a powerful opt-in

Would love to know your thoughts on this feature Thanks and have a good one!

[slavakurilyak]

@fire17 are you expecting the project's sponsors to cover Pinecone costs? It's important to clarify this aspect to ensure proper resource allocation and support for the feature.

[Boostrix]

This is basically discussing the same idea:

• #1749

One idea being to use a separate "proxy LLM" in between auto-gpt installs and the openAI endpoint to gather a ton of data that way, and come up with an intermediate model: #348

[github-actions[bot]]

This issue was closed automatically because it has been stale for 10 days with no activity.