They're not really reversing anything as much as they're making their system straight up refuse with ML tasks.
But the problem is that Anthropic showed their hand. They showed that they can use steering vectors to degrade outputs on specific subjects.
That's why people using DS4F with DwarfStart, 2 bit quantized, are often surprised by the results. It's not a frontier model but it is not a toy, it is something you can actually use to get work done, and nobody can tell you want to do with it.
Key 🗝 🦊✅
antirez
idk if I can call myself "active" after throwing one half assed project at github, but in any case I hereby resign from the community of vertebrates as I caved in and restarted my sub just to talk to Fable while I still can
I won't mourn OAI if they decide to fuck off and die, but if they can't keep up this sets a bad precedent. I don't want a world where a single US company controls what anyone can do with the world's strongest computers, please tell me that this isn't happening.
"amateur number engineer dead in combinatorial explosion"
I sure hope this will be worth it.
Gemma 4 now runs 2x faster with MTP GGUFs! Run locally on just 6GB RAM. ⚡️
MTP enables Google Gemma 4 run ~1.4–2.2× faster with no accuracy loss.
Gemma 4 12B MTP can run at 162 t/s vs. 52 t/s without MTP. 31B reaches 101 t/s.
GGUFs + Guide: unsloth.ai/docs/models/...
this is interesting, i'm v curious to see how their memory system works out huggingface.co/silx-ai/Quas...
gods I hate how people are now trying to turn "transhumanism" into a snarl word. "the scary evil billionaires are trying to brainwash our children to make them accept unnatural human enhancement" fuck entirely off with this shit
Still in need of a full-time remote senior/staff-level job!
My strengths are TypeScript, Solidity, SQL, C#, and Swift. Decent with Rust. I'm a good team lead too and was once a CTO of a YC company.
Boosts for visibility appreciated. Thanks SO much.
