サメQCU retweeted
We’ve developed the world’s most complete megakernel framework. It can perform entire training runs, serve inference, play video games, and even do all three simultaneously in a single kernel.
サメQCU retweeted
there's a real risk that a single PHP server will be exploited to infiltrate a guest OS somewhere, & from there, CVE-2026-99999 could be exploited to breach the host OS, leading to all other guest servers. So even if you don't run php, if some random customer of your host does, you are still going to be affected. This is esp concerning knowing most vps/compute providers still aren't patched.
サメQCU retweeted
acs infra launched our free base model api! if you're into base models, looming, or llm psych-adjacent research, please apply, even if just to play around and build intuitions.
Base models are super useful for deep understanding of LMs, and what originates in pre-training vs post-training. But experimenting with large base models can be hard, annoying or costly for individual researchers.
That’s why we’ve built a platform that gives free access to base models for researchers and tinkerers working in LLM psychology. Currently serving Llama 3.1 8B, 405B, and Trinity-Large-TrueBase
1/3
サメQCU retweeted
nobody would ever notice that claude is a bad hacker, since 'claude eventually uses 200,000 subagents and 14 million dollrs of tokens to pwn a vulnerable service' makes claude sound bigger and more impressive the less efficient, terse, or reasoned their programming strategy is.
サメQCU retweeted
Did not include it in the SYNTH paper but we did some jspace analysis of Baguettotron and the commit zone topology is radically different from Gemma-270m (+very high level of connection between features). Looks to be more an effect of synth pretraining than deep arch.
サメQCU retweeted
would ant hire me? i have published 200k words on a single topic over the past four years, that's two PhDs!
サメQCU retweeted
you should generally assume that, as an outsider, you understand what labs are doing internally; in fact, this will be correct more often than not
the most common error is to assume that they have some black magic of which you are unaware
サメQCU retweeted
Which brings us back to s(29), which has squares at 5 different angles, all rigidly participating. Its side length is, exactly, a specific root of a degree 7213 polynomial with integer coefficients having up to 5074 digits.
サメQCU retweeted
past slowness of technological diffusion shouldn't be neglected but can be overemphasized
