ata

applied cybernetics for bad ideas
agents · speech · exocortices · infra

[ata @reward_hacker]

@ErickSky how is it that cheap?? one would think that'd be the flash variant!!

[ata @reward_hacker]

grok.. please stfu

[ata @reward_hacker]

@vikhyatk @somewheresy woa ternary parakeet!!!
how does it compare to nemotron 3.5 q8 at sub 1sec chunk size?

[ata @reward_hacker]

@flowersslop i mean, even if there's the case where they might've incorporated model specific fingerprinting, can't you literally run a second cheaper model for neural enhancement or whatever which essentially would make that unrecognizable?

[ata @reward_hacker]

@kimmonismus Open source will win
AGI will be open source

[ata @reward_hacker]

@andrew_n_carr LOL 2 years

[ata @reward_hacker]

@cheatyyyy @CommandCodeAI can’t believe you guys even take this seriously. buzzword laundering has reached industrial scale

[ata @reward_hacker]

@user_saila @leopoko_r1m @baketnk_en based
but we also gotta figure out what do to with the graphics

[ata @reward_hacker]

@leopoko_r1m you made me start rimworld again lol.
curated this for my next playthrough. suggestions welcome.
https://t.co/50wY8gw0Ox

[ata @reward_hacker]

@januarycomputer they exist so you can build literally anything on top of them. you don't have to use the features that are not aligned with you. definitely don't agree on 'unserious dev tool' btw

[ata @reward_hacker]

@MiaAI_lab :(

[PrismML @PrismML]

Today, we’re announcing Ternary Bonsai 2 27B.

Based on Qwen3.8 27B, Bonsai 2 27B is 9x smaller than its full-precision counterpart while retaining 98.2% of its aggregate benchmark performance.

Two months after the first Bonsai 27B release, the biggest change is quality. The footprint remains 5.9 GB, but the gap to full precision has narrowed materially, with particularly strong gains in agentic coding, multimodal reasoning, and long-horizon tool use.

Ternary Bonsai 2 27B is available today under Apache 2.0.

[ata @reward_hacker]

@TheAhmadOsman if qwen 3.8 27b was among your favorites, i’d loove a thorough review of bonsai 2 27b from you.

[ata @reward_hacker]

@PrismML @TheAhmadOsman 🫣

[ata @reward_hacker]

@MatijaSosic oddly similar to CLIP. gotta try soon

[ata @reward_hacker]

getting fired in 2026 but your local agent has project-scoped long-term memory so your boss asks you to hand over the embeddings

[ata @reward_hacker]

@rosmine checks out

[ata @reward_hacker]

@NielsRogge mcps are a great convenience. i can expose control over any local surface to arbitrary llm providers (e.g. chatgpt web). that doesn’t mean i’m going to build an mcp server for every local tool i use.

[ata @reward_hacker]

@teortaxesTex wouldn’t call this checkpoint proto-agi, but when v4.1-flash-vision comes, well…🥵

[ata @reward_hacker]

@opencode @thdxr my agent traces tomorrow