we did something similar at @splits to detect NFT spam more effectively using LLMs. worked really well.

Trending topics on Farcaster took less than a week to build. 

We use LLMs to categorize each post with some prompting, and then use a rule based system on top to discover topics.

It's surprisingly effective, and took about 1/10th of the time of a traditional ML approach. https://t.co/9SNaADm4u6

Varun Srinivasan

splits.org

active at /splits, come say hi 

we did something similar at @splits to detect NFT spam more effectively using LLMs. worked really well. https://x.com/varunsrin/status/1925342791191994630?s=46

design @splits | /wireframe /latenightcrew /event-pass /feud

Michelada connoisseur. Chief Onchain Officer at /area51 since 2022 AD. Focused on digital art, tech & crypto. https://gramajo.carrd.co

i think stuff like this highlights what is most underrated about the models and what they can do: repeatable workflows where there is judgement or probability involved

I/legal manage our external facing emails for asana and the amount of shit that comes out way is bonkers. But we still have to sort it for the actual real things.

With Claude that email and triaging is on autopilot 🤣 it’s kind of nutty. So many hours saved