Chiming Wang retweeted
Crypto grew up in the Dark Forest, and the entire world is about to find itself there
> be humanity
> spend decades building the most autistic tank shells imaginable
> chemical explosives, shaped charges, electronic detonators
> then everybody just converges on sabot rounds
> long dart yeeted at 1700 m/s
> best way to kill is still just throwing long sticks
why are so many countries trying to develop unjammable AI guidance for drones when the technology was already invented 8 decades ago ??
capable uncensored models should be easily accessible to everyone, not just as gated FP8 weights.
i've released GGUF quants for OrcaRouter's abliterated Qwen3.8 27b: uncensored opus 4.5-level capability on a single GPU :)
huggingface.co/chimingw/Qwen…
We just released our uncensored weights for Qwen3.8 27B FP8, designed for AI Red Teaming and security research, enjoy!
huggingface.co/orcarouter/Qw…
what happened to Tom Jobbins? i think his mistral 7b v0.1 was THE first model i ran locally
he was the best back in 2023 :(
new deepseek v4 flash and qwen3.8 27b are going to change everything
Replying to @cj_pais
it probably will not take that long (few years?) to get to ~50 tps on a small Mythos-level model; the small local models (for laptop and phone) today are better than what was considered frontier-level just 3 years ago
qwen3.5 9b even outscores gpt-4.1 and gpt-4o (1t+ ??):
war is just chess but you can just give your pawns FPV kamikaze drones and instantly upgrade them to rooks at any moment so it's insane
America is destined to win and nothing can stop us
This is gonna sound weird: but I kinda believe in a version of the anthropic principle that says America produced so much wealth and population growth globally that any conscious being is disproportionately likely to have emerged in a universe where American was founded, so certain inexplicable events happened specifically to bring America into existence.
Chiming Wang retweeted
Replying to @cj_pais
it probably will not take that long (few years?) to get to ~50 tps on a small Mythos-level model; the small local models (for laptop and phone) today are better than what was considered frontier-level just 3 years ago
qwen3.5 9b even outscores gpt-4.1 and gpt-4o (1t+ ??):
there is a very real possibility that both the US and Chinese governments make it illegal for AI labs to release open source models
me in 2027 going to Wyoming because Kimi K3 is above Mythos-level and GPUs are banned in California so i have to buy RTX6090s at the underground compute markets
soon governments will restrict your access to compute
ID verification for buying and running GPUs is coming
or they’ll mandate systems to detect if you’re running banned AI models, similar to 3D printing in New York (ghost gun printing)




















