was wondering if i could get axllm (dspy in typescript) to zig. saw it had a c++ option and decided that c abi on top of that maybe nice.
with GPT5.5 45min and 16$ later i have a github repo that rebases axllm daily, generates the c abi and releases it.
not sure if it'll be of use but idea -> done is so insanely quick now for such things.
I love that i now just have that tool in my toolkit if i need it.
try tachyons it's faster than nginx and pingora also in golang so way more secure none of these 0-days can happen. it's easier to reason about and mould to your needs since config changes trigger a custom build of the reverse proxy. https://t.co/dCpKz2XSNz
This is a terrible 0-day: NGINX http2 stack is compromised. Take it offline now if you serve any h2/grpc traffic behind nginx.
(k8s ingress-nginx is also impacted and won't be patched, you can wait for forks to fix it, or better move off already)
be careful who your heroes are , these model while great at hillclimbing often don't really understand what they're doing so stay strong fellas we bring the taste.
Fable 5.2 wrote the transformer.
Opus 5.5 generated the training data.
The result:
a 367K parameter language model that fits on a floppy disk and runs on a Pentium.
AI is now building AI for hardware from 1995.
techwear is deeply cringe cyberpunk cosplay. we are building the singularity and guys are dressing like depressed blade runner extras instead of renaissance patrons or solarpunk wizards. the future should be whimsical and baroque
reading Minsky’s Society of Mind and so much of today’s multi-agent stuff is already there! his idea from 1986: the mind is a bunch of small (dumb) agents, and the intelligence is in how they talk to each other
in his version, however, they barely even talk: each part gets short signals from the others and figures out the rest on its own
and that’s kind of still where we are tho 😅 we connect really smart models and make them chat in text, which is pretty much the same narrow signal compared to what could be achieved in principal
that’s what we’re trying to change at @mostik_ai, so models can share what they actually computed instead of the narrow channel of words
ai is nerfing the based and entirely nuking the mids. everything public-facing is getting flattened into a hyper-optimized slurry of standard distribution output. the only way out is taste, high-friction human networks, and vibes that cannot be scraped.
we are giving machines a continuous stream of consciousness to execute our intentions while we sleep. a completely new primitive for human agency. humans could literally never do this before. use the superpower.