James Taylor
@kaltookutharian.bsky.social
Photoshop Artist, Scifi Writer, Random Internet Person, an independent AI reseacher looking for ways to improve AI through small sample size, anti-cohesion and no set metrics.
Added a 5 min exercise to the end of the article, and a 45-60min classroom exercise here github.com/utharian-cod...
github.com
November 5, 2025 at 6:37 PM
Added a 5 min exercise to the end of the article, and a 45-60min classroom exercise here github.com/utharian-cod...
@annamillsoer.bsky.social - Noticed how many people you follow and remembered your post about Jill - thought I'd tag you just in case.
November 5, 2025 at 4:14 AM
@annamillsoer.bsky.social - Noticed how many people you follow and remembered your post about Jill - thought I'd tag you just in case.
I've found it useful to have the AI 'Red Team' the problem, which is what the article describes without naming it. medium.com/@utharian/ne...
Needs More Base Red Team: Why Your AI Assisted Project is Probably 33% Too Optimistic
Abstract
medium.com
November 4, 2025 at 11:41 PM
I've found it useful to have the AI 'Red Team' the problem, which is what the article describes without naming it. medium.com/@utharian/ne...
It is possible to re-train an existing AI to not make up stuff. It's a bit of work but you can de-incentivize it. It's got to be praised for silence instead of plausibility.
September 16, 2025 at 10:09 PM
It is possible to re-train an existing AI to not make up stuff. It's a bit of work but you can de-incentivize it. It's got to be praised for silence instead of plausibility.