this is better than TFA

sometimes i wonder if the left hand knows what the right is doing. it looks like we arrested our own spy in this case: https://www.politico.com/news/2026/05/25/american-journalist...

The CIA director requested the FBI intervene. This is almost certainly not a fuckup.

This could also be internal politics intentional designed to burn someone for pissing off the wrong people. That shit happens.

That's their post hoc, uncorroborated claim. It's easy to imagine many other possibilities; it could just be face saving. It could be Rush is taking the fall. etc.

absolutely. somebody online was wanting an LLM with Georgian language support, and that's exactly what i suggested: start digitizing Georgian text.

i'll upvote this each time it's submitted

you should be using dflash with that model, look it up

p-e-w was just talking about this the other day in his Discord. seems doing the one neuron method is quite bad for KLD and that's why the newer techniques have stuck.

Who is p-e-w? Is it a public discord?

heretic maintainer: https://github.com/p-e-w/heretic

the fun bits are in another branch or PRs

site has so little information there doesn't seem to be much to discuss

There seems to be some additional information in a video on the creator's YouTube page:

https://www.youtube.com/watch?v=cPFmcVtGnh0

edit: There appears to be more information in a variety of the creator's videos, but it looks like the projection from birdsong-to-3D is (probably simplified) taking the Mel spectrogram (40 features from this, unclear as to what they are) and passing it to PCA to get 3D vectors.

The creator discusses a real-time version here: https://www.youtube.com/watch?v=prhQqpAxrm8

this looks awesome. i've been struggling with vector compression, and have been trying PCA + all sorts of rotations. looking forward to trying this out

nice writeup! looking forward to doing some more training as soon as i get some more data sorted. it'll be a custom arch, but i'll probably shoehorn it into unsloth for a speed boost.

Thank you!

you can train it, but not fully

I trained karpathys d28 1.6b nanochat on a 3090. Took an extremely long time but I did it.