TheExGenesis / community-archive

An open tweet database and API anyone can build on.
https://www.community-archive.org
MIT License
54 stars 8 forks source link

Short blog post on the community trends app #157

Open DefenderOfBasic opened 3 weeks ago

DefenderOfBasic commented 3 weeks ago

This would be useful for showcasing a concrete, useful use case of the twitter community archive. And also as inspiration for developers.

General ideas / outline:

DefenderOfBasic commented 3 weeks ago

Defender's thread, "science vs spirituality" https://x.com/DefenderOfBasic/status/1846905536115024307

Comparing ChatGPT against google trends https://x.com/DefenderOfBasic/status/1847656217427984649

Xiq's original post: https://x.com/exgenesis/status/1846578431212802156

(this is more general but want to put it somewhere for now: Xiq's thread of app ideas/stuff people are building on top of community archive https://x.com/exgenesis/status/1835411943735140798 )

DefenderOfBasic commented 3 weeks ago

Maybe the BIGGEST example is looking at the release of ChatGPT, and comparing that with google trends?

DefenderOfBasic commented 3 weeks ago

Looking at the very first usages of the word "LLM" in the archive is fascinating. Here is 2021. It'd be interesting to see who was getting it right early on, and if their reasons make sense today, and those would be very good people to follow/hire/consult?

Papers of LLMs are barking on the wrong tree. If the main effort was to get all the right parameters and engineering efforts. Why doesn't it take most of the report? This is the main thing we can learn about the next LLM.

https://x.com/i/web/status/1473608713625812996

(ChatGPT release date is November 30, 2022. The "attention is all you need" paper was 2017?)

DefenderOfBasic commented 3 weeks ago

Twitter recently released a "radar" premium feature that lets you do this kind of trend analysis on real-time twitter data. This is validating that there's real value here. The "radar" tool only lets you search data up to 7 days ago. Also we can't improve it/add other data sources/do semantic search