a holder count that treats every wallet as a different person never sat right with me.
so I built nuno: it groups linked wallets behind https://t.co/Q3PwIHxtzf tokens into real entities and shows real holder counts + organic volume.
open API, on @monad 🔍
Fable 5.2 might require less context to carry out tasks in exchange for incredible token consumption
But, that doesn't define the output quality
Expensive ≠ better
no big commits today, just thinking. the best part of mene is a lockfile check that stays quiet until something is actually wrong.
That's harder to design than loud.
tomorrow i might tighten how it flags hijacked npm packages. https://t.co/1T7yf2HhwS
fine-tuning can quietly wreck a model's safety, so I built brink: it gates risky training data, watches refusal while it trains, and rolls back before things break. this morning I added cost per 1,000 replies vs closed APIs, plus the write-up. https://t.co/BVNlUjCme1
long sunday on brink, my tool that stops fine-tuning from breaking a model's safety.
ran all six stages, put real numbers in the write-up, made the @huggingface space public. pipeline finished, then the last circuit step failed 🙃 that's tomorrow
https://t.co/WoKH6lJCHX
a holder count that treats every wallet as a different person never sat right with me.
so I built nuno: it groups linked wallets behind https://t.co/Q3PwIHxtzf tokens into real entities and shows real holder counts + organic volume.
open API, on @monad 🔍
One private wallet funded 10 wallets in 17 seconds.
Those 10 wallets launched 102 of the 426 @naddotfun tokens from Oct 4–14.
Nuno finds the real owners behind the wallets, with a Monad transaction for every link.
Count people, not wallets on @monad.
https://t.co/U2RMnmD42b
Fable 5.2 might require less context to carry out tasks in exchange for incredible token consumption
But, that doesn't define the output quality
Expensive ≠ better
Agents tend to do surface level work regardless of model, each model just completes the task
More efficiently than the previous, yet to see one that understand a concept in its whole entirety
And performs in accordance with less guidance or provision of more context.
no big commits today, just thinking. the best part of mene is a lockfile check that stays quiet until something is actually wrong.
That's harder to design than loud.
tomorrow i might tighten how it flags hijacked npm packages. https://t.co/1T7yf2HhwS
>npm warnings live on websites anyone with the right password can edit or delete.
>
> Mene puts them on @Arkiv_Network: sealed so not even the author can delete them, confirmed by 2 of 3 independent witnesses, readable with no server.
>
Last week I needed to download an Ig vid so I built a bot that will help with that
Meet WTFSOCIALBOT
Over the weeks I started using the bot to solve some of my common problems with new features
Today I made it public so at least I'd help out with some people with their lil issues for Free
fine-tuning can quietly wreck a model's safety, so I built brink: it gates risky training data, watches refusal while it trains, and rolls back before things break. this morning I added cost per 1,000 replies vs closed APIs, plus the write-up. https://t.co/BVNlUjCme1
long sunday on brink, my tool that stops fine-tuning from breaking a model's safety.
ran all six stages, put real numbers in the write-up, made the @huggingface space public. pipeline finished, then the last circuit step failed 🙃 that's tomorrow
https://t.co/WoKH6lJCHX
long sunday on brink, my tool that stops fine-tuning from breaking a model's safety.
ran all six stages, put real numbers in the write-up, made the @huggingface space public. pipeline finished, then the last circuit step failed 🙃 that's tomorrow
https://t.co/WoKH6lJCHX
kept thinking about how a hijacked npm package can just sit in your lockfile for weeks and nobody notices
I built Mene for the Arkiv hackathon. it scans your lockfile and flags them, and the advisories live on Arkiv so nobody can quietly edit them
Unfortunately I didn't win.