As usual, classifiers seem incoherent. Opus 5.5: Can you create me a flash attention kernel for... ✅ NVIDIA B300 ✅ AMD MI355X ✅ TPU7X ❌ Huawei 950DT ❌ Trainium3 But that means...

Yeah so a quick test suggests anthropic is targeting Chinese hardware with their classifiers. Someone with some more time should do some classifiers probing, but it seems like Dario's grudge match with China continues.

this is something i hadn't taken seriously enough. it's not wrong. if backdoors get deep enough in the supply chain it can be hard to be sure you ever got rid of them. we should, for example, be spending special effort on securing all systems related to building compilers

UNDISCLOSED PAID MEDIA: The salaries of reporters, Harry Booth and Billy Perrigo, were paid by Doomsday cultist Dustin Moskovitz's foundation Coefficient Giving via the Tarbell Fellowship.

I am in a minor compute emergency and looking for 128 B300s for a month that we could bring online ASAP. This week ideally. Next week at the latest. Is there any world in which this is possible? Assume money is not the constraint Text me if you have my digits DM me if you don't

I’ve built DNA synthesizers and sequencers by hand. I used the engineer viruses for a living. I used to engineer human immune evasion for therapeutic constructs. Who the fuck are you people? Have you ever so much as held a pipette before? You can spaghetti blast an ensemble of DNA sequences at some shitty provider but 1) you still have to assemble it and bootstrap a system for making virions 2) you are not single shotting a viable, virulent viral design without a ton of experimental selection and development. Viral fitness is deeply dependent on codons and cotranslational kinetics - you’re not just gonna obfuscate away from wild type and get something good by magic. You keep treating AI like some kinda god, but molecular physics has computational complexity that scales exponentially in particle number which just crushes the abilities of any classical computer to do end-to-end design of biological functions ab initio. Grabbing a bunch of bacteriophage phi174 hits from a mass ensemble screen in lab microbes is not evidence of some magical AGI bio design ability - it's just a classic spray and pray selection. This is just nothing like building something viable in humans. Goddamn it read some books before you waltz into biomedicine and lecture us on protecting human life.

why would anyone call consider "browsing the internet to look for solutions" an example of "escaping containment"? what "containment" are we talking about? this deserves a byline check! oh, look...

Good post, but I have to nitpick this thought experiment, even if I agree with the larger point. The 20 years ago version of you is not best equipped to evaluate headline facts in context, and likely brings priors about what those facts imply that are simply not true.

We're partnering with @boltdotnew on Forge, a new initiative to help train our next-generation open models using real-world development data. Starting today, eligible Bolt Pro users who opt in get 50x more usage across top open-weight models, including DeepSeek-v4-Pro, GLM-5.3, and Kimi-K3 with no daily caps. In exchange, anonymized build sessions will directly inform Arcee's model training and evaluations. Once the training wraps up, these models will be released with weights published and freely available to the public. Learn more:

@yacineMTB I was struggling with an inability to disconnect recently too. I wrote up my thoughts, curious if it resonates. Basically, our sense of how to spend our time is becoming miscalibrated, because things have changed faster than our measure of value adapts.