Ever since the Scale.ai hackathon, i've been thinking more about GPT as a backend.

The most practical usecase for me is mocking out apis while I build my React UI.

I like react-query, so I hooked it up to Promptable.js (Langchain.js in TS)

and built GPT-QUERY: What's it do?

Just pass in a zod type and get a query response that includes real data generated by GPT-3!

->

Are you a NOOB Prompt Engineer, failing to summon the power of the GPT?

Do you look to @goodside for guidance?

Look no further:

Introducing Promptable.ai

The Ultimate Workspace for Prompt Engineering 🪄 First things first, Promptable has been 🙏 blessed by
@HailGPT
with heavenly completions.

Using GPT-3 and getting bad results?

Hop on over to Promptable.ai.

Pull out your old crusty Prompts and watch as they magically work first try! 😉

Anyone else trying to use EXTERNAL knowledge (databases) in their GPT-3 prompts?

e.g. adding ‘context’ to a prompt or chain to generate better results?

just saw this paper “Rethinking with Retrieval” which looks promising.
arxiv.org/abs/2301.00303

A short breakdown: TLDR
⁃LLM’s don’t have complete knowledge (yet)
⁃This can lead them to predict wrong answers / invalid facts etc.
⁃So we need a way to tell them about recent events / new data
⁃Combined with CoT/self consistency, this leads to better results!

How?