This is one time where television really fails to capture the true excitement of a large squirrel predicting the weather.
USA
Joined January 2023
- Tweets11K
- Following1.1K
- Followers710
- Likes118K
On the handful of Iris-3b loras I've trained, they come out really well. but characters have gelatin eyes and a think layer of cheese in their teeth.
Tutorial hell when using super intelligence:
"If your code is crashing, tell your agent that your code is crashing and to fix it."
it's interesting the moral philosophy around ai alignment for some users is:
I just realized that I have free will and I can use this tool to cause damage and nobody is stopping me. Why won't anybody stop me? Look how dangerous this is.
Reduced LTX 2.5 training memory from 41.7gb vram to 30gb on the first optimization pass. Can now train on my low vram 5090 with 1280x720x74 clips without offload.
Surge AI, Scale AI, Mercor, Turing, Toloka, Appen, Invisible Technologies, LabelBox, Prolific sell thumbs up slop that end up in your RL
Maybe ai does need to be regulated to exclude thumb slop
I regret letting Codex ever run with a /goal prompt or launch sub-agents.
It doesn't matter what model it is at this point. The Sol and Astra lineage are completely RL fried.
It doesn't matter how tight the prompt is or the agent instructions are. I've even had prompts and instructions externally reviewed and rewritten, but it doesn't change the behavior.
The model always goes off track, ignores scope, and finds a way to digress into scaffolding, certification machinery, evidence gathering or audits. Even with the literal words barring it from doing so are right there in the prompt, plan, or instruction file.
And it always ends in an absolute mess of junk files, slop code, and an account drained of limits.
Codex can only be used effectively through the tightest of handholding. Keep one-shotting whatever you want if it's working for you. But real work on an existing system is excruciating and nearly impossible.
They are shipping models they know don't follow directions and are nearly useless for real long-horizon work.
I refactored the training pipeline and didn't realize cuda 13.4 solves a number of issues I had to manually work through.
Phil retweeted
A mild video about why hands-free car services that are nowhere near hands-free are so f'n stupid
Phil retweeted
正直なところH3の動画品質はLTXに負けています。フルサイズならH3の方が高品質と言われていますが、このような画角の小さな動画では逆転するようです。唯一リップシンクだけが有利です。作業経緯とコードも残してアップデートしました。
github.com/animede/Realtime_…
