
The Community Archive maps how ideas spread in online communities! community-archive.org/
✓Mapping how ideas spread in online communities

The Community Archive maps how ideas spread in online communities! community-archive.org/

Inspired by @edwardjyoung's LW post "Our most aligned model yet", I was curious to check whether the alignment benchmarks reported by the labs can be backed up by measures of real-world misbehavior. And what better source of real-world behavior reports is there than Twitter! If Fable 5.2 has an misaligned behavior score of 1, we can (imperfectly) extrapolate the score of each Anthropic model since Sonnet 4.5, as each new model that comes out reports comparisons with recent models in the system cards. And we classify tweets in the @comm_archive that mention each model as whether it is a first-hand report of it misbehaving, excluding retweets, jokes, rumors, and reactions to benchmarks. This is just the v1 with lots to improve upon! But yeah, currently uninclined to believe the labs when they announce the next "most aligned model yet."

If you want more time series, you can check out the new charts at modelbehavior.report Discussions about claude crush the other model families

🌊 RIDE THE MODEL WAVES 🌊 % of active users in the @comm_archive mentioning a model (~ 1k active users at any point in time). Opus 5.5 enters the pantheon along with GPT-4, Deepseek R1, GPT-5, and Mythos

guys I made it I'm officially an ML researcher as approved by the @comm_archive

New @comm_archive front page. Opt in to find where you are in the map! I need new 43 users before demo day 👀 community-archive.org/

🚢🐦 COMMUNITY ARCHIVE BUILDERS 🐦🚢 @comm_archive is moving to a new API and deprecating the old one in 30 days Please send your coding agent our docs and update community-archive.org/docs

🚢🐦 COMMUNITY ARCHIVE BUILDERS 🐦🚢 @comm_archive is moving to a new API and deprecating the old one in 30 days Please send your coding agent our docs and update community-archive.org/docs

was just chatting with @maskys_ about @comm_archive and how maybe it could help facilitate something like this; like a proto-publishing house (comm_pub?) that surfaces the proto-essays and books latent in people’s tweets then lets other people discuss and encourage

this is so cool! Delvetown agent trained on @comm_archive data. I will have to get deeperfates running soon so they can hang out 😈

chatting with Alexandre was a main inspiration for me for making goodbots.review! check out his cool site, made with @comm_archive ✨

omg new research! (on @comm_archive data)


built a tiny app upon @comm_archive, that shows you your first tweets (link below)

hello world!