shoni.eth

@alexpaden

616 Following

37255 Followers

shoni.eth pfp

this is another app that needs a non-existent data layer which, in every other networking app, comes from a survey. It’s a social ai problem. Here’s 100 people at the event, who should I meet?

0 reply

0 recast

0 reaction

shoni.eth pfp

0 reply

0 recast

6 reactions

shoni.eth pfp

i don't know if farcaster data is or ever will be valuable-- but i do know that - social sequences are actions taken by humans to achieve a goal - grok does not know who i am, regardless of the profile button - farcaster might be all social networks - crypto enables decentralized data access, model hosting, and even training opportunities - data will grow and become more generic and less valuable this is why i am still excited about it, whether i put fc data down for months or not, i know the opportunity will continue to flourish.. and when people approach social ai simply from the mindset of search automation, that's the plot of walking dead.

3 replies

3 recasts

15 reactions

shoni.eth pfp

oh i'm shocked that people who invest life savings into internet memes have a bad social environment

0 reply

1 recast

9 reactions

shoni.eth pfp

been thinking about this one since Elon said grok was the new feed alg- lots of interesting approaches. One of them is the profiles dataset which acts as a map for personal interests— follow the leader ai are low effort and novelty, this almost always turns to gimmick after a couple uses. this concept is now dead imo, but the map is still partially alive.

0 reply

0 recast

4 reactions

shoni.eth pfp

1 like gang gang https://farcaster.xyz/alexpaden/0x702d47be

0 reply

0 recast

4 reactions

shoni.eth pfp

trying something new for a bit and I’ll be out of america for the duration. dataset releases and novel farcaster ai research will continue 🫡

0 reply

0 recast

3 reactions

shoni.eth pfp

ion have much to say anymore just be trying to grow as a person and make friends that appreciate the person. feels like looking for the invisible key in a dark room suppose by the time it’s found that feeling seized to exist “This is your moment, and every single minute you spend tryin’ to hold on to it ’cause you may never get it again…” - till i collapse

2 replies

1 recast

18 reactions

shoni.eth pfp

when did the Silicon Valley advice change from take risks in your 20s to start out working for one of the big companies

3 replies

0 recast

6 reactions

shoni.eth pfp

farcaster pro users be like I thought you’d want to read that opinion in long form

1 reply

0 recast

9 reactions

shoni.eth pfp

normally I’d have some stuff to say about the autofollow list but these days I just want some horses on a ranch

5 replies

0 recast

15 reactions

shoni.eth pfp

New Release: https://huggingface.co/datasets/shoni/farcaster Total Records: ~18,147,313 threads Data Cutoff: 2025-06-25 (no threads newer than this date) Notes: Includes f32 and binary embeddings, new thread formatting, recursive quote hydration, and more efficient token inclusion

2 replies

0 recast

8 reactions

shoni.eth pfp

~18M unique threads in threads/v2 dataset-- pgvector to parquet is slower than expected.

0 reply

0 recast

7 reactions

shoni.eth pfp

cursor replaced $200 gpt for me.. number go up slowly

1 reply

0 recast

5 reactions

shoni.eth pfp

[this is an automated test please ignore] ill use likes/recasts from this post as a trench qualifier for airdrop v1

4 replies

9 recasts

43 reactions

shoni.eth pfp

threads/v2 was 2.148(B) tokens per embedding completion primary growth from Quote hydration, and efficient token context

0 reply

0 recast

3 reactions

shoni.eth pfp

quote with your drafts

2 replies

0 recast

8 reactions

shoni.eth pfp

"This is an automated test reply." before you can know what to say, you must know where to say.

0 reply

0 recast

3 reactions

shoni.eth pfp

0 reply

0 recast

3 reactions

shoni.eth pfp

i think potentially a useful miniapp is user search so i'll do that mini-PRD: - search by prompt - result list and profile view searchlite first claude code project, i guess the goal is keeping capex low. release after profile v1 dataset, no public v2. (this is context)

0 reply

0 recast

4 reactions