A public good that would be really useful is a public dataset of embeddings for casts. 

Assuming it's legal to do, would drastically reduce the cost of building embeddings based models on Farcaster

Technowatermelon. Elder Millenial. Building Farcaster. 

nf.td/varun

like get a cast, it's embeds, and return head og tags for each embed (url)?

cto @ bwl.gg (lunchbreak.com), 80M+ users (bdut.ch) 🙏 built @remindme, @mintit, @gpt, @healthbot, farcantasy.xyz

no, i mean like vector embeddings for the text

What are you using the embeddings for in training- it's typically search and classification 

llm fine tuning and new training uses raw text

i can provide snapshots if someone would like but custom transformations then embedded are where the value would be.. embedding every raw cast is a lot less useful imo