wizard not parzival
@alexpaden
the gap between old people (trad docs) and young ones (llm first) is wild here ngl.. without dtech i had 0 primer for building actions on fc
1 reply
0 recast
2 reactions
wizard not parzival
@alexpaden
cc @rish this is way more important and flexible for me than custom gpt fwiw.. custom gpt is 4o only
1 reply
0 recast
0 reaction
rish
@rish
does it need to be .txt? Our API spec is already a single .yaml file: https://github.com/neynarxyz/OAS/blob/main/src/v2/spec.yaml
2 replies
0 recast
0 reaction
rish
@rish
@dylsteck.eth @kevinoconnell curious what you think would be useful
1 reply
0 recast
1 reaction
dylan
@dylsteck.eth
i wonder how some of the startups that are creating these .txt files automatically generate it but i do think either something like would be cool -- i'll also have to experiment with how good gpt is with a flat .yaml file because honestly maybe that's all the context it needs 🤷♂️
1 reply
0 recast
0 reaction
wizard not parzival
@alexpaden
re startups doing it: feed in the api and sample results to get descriptions. do the same for high level sections (i.e. cast or user related apis). put them all together including api curl, result sample, and description. seems like other libs would be a pain. if you went totally crazy next level on this, it would look something like this for state of the art llms https://blog.alexpaden.tech/online-identity-as-an-ai-crypto-agent-prompt-framework-t1 md for llms -- the big thing is i wanna spin up a sample of something very quickly, it needs a little primer and ideas by default... it already knows i.e. next.js but it needs to understand how people think about neynar apis and it can use sampler data to its advantage (is what i found from column names vs first row results in query gpt). context is usually 100K at this point, and more importantly in-chat embed chunking and semantic search isn't that hard.. that's part of what the chat interface is doing and why long docs are now more acceptable
1 reply
0 recast
0 reaction
