7777777phil

The entire screening infrastructure probably assumes cardiac events are a 60+ disease. Younger patients get atypical presentations, get sent home with "anxiety," and die. Metabolic syndrome …

Makes sense. This also tracks with the research on human-AI collaboration. A single model converges to the mean of its training distribution, but adversarial multi-model setups break that pattern beca…

562 total California test miles since 2016, (also vs Waymo's 13 million) FSD does put miles on public roads but none of them count in California's regulatory ladder because they're all …

The Videogamer thing is a good illustration of why AI specifically fails at criticism. Next-token prediction and RLHF optimize for typicality, so you end up with text that reads like an average of all…

Amodei is probablyright that current models aren't reliable enough for high-stakes decisions, but the more useful question is what the failure mode looks like in practice.

the article is framing this as DeepSeek hiding sanctions violations but it's probably simpler. Give Huawei early access before Nvidia can optimize for the model and you help build a domestic chip…

Sorry, this post has been removed by the moderators of r/LocalLLaMA. Classic reddit..

32B model in 19.3GB matters is really cool imo. Memory and cold start are what gate production deployments. I did a piece (1) on how Netflix and Spotify worked this out a while ago, cheap classical me…

so 44 million tons of CO2 by 2030 from new gas turbines alone, and coal plants coming back online on top of that. Meta building a 5GW gas-powered campus while putting out sustainability reports is... …

Carney at Davos a month ago, almost word for word: >middle powers need to stop pretending the rules-based order works and start building sovereign partnerships. Wrote about it in Jan: https:/…

AWS's answer to an AI-caused outage is mandatory peer review for AI-generated changes. So the solution is human oversight, the same humans they've been cutting by the thousands. They took th…

15% defaults would be brutal but honestly not that surprising. AQR's 2026 capital market assumptions (1) had U.S. buyout returns at 4.2% vs 3.9% for public equities. A 30bp premium for years of l…

sing the "supply chain risk" designation against a domestic AI company is wild. Not sure that tool had vendors who won't rewrite their ToS on demand in mind. Meanwhile the Pentagon coul…

128 to 210 WEEKS for power transformers. You can fab more GPUs in 18 months but the electrical infrastructure to run them takes 3-4 years and there are maybe a handful of manufacturers. Feels like the…

NIST asking for agent security comments right as the agent stack is splitting into layers with completely different threat models. A model layer vulnerability looks nothing like a tool-use layer vulne…

thanks for getting back to me, very cool if true :) I have been asked about this many times when talking LLM use cases at enterprise level. Would love to run som tests, pleas shoot me a message to the…

The token comparison is what jumped at me.. Half the context window for the same work means mainstream tools are burning your tokens on system prompts and MCP plumbing before you even start. I wrote e…

If this decomposition actually holds, it's the first model where you could show a regulator why it produced a given output.

Props to Tao for his response: he didn't just patch the sign error, he went back to Hildebrand's paper and found a better fix through log-concavity of the Dickman function.

Odd strategy (I guess at this point nothing really is with Musk / Tesla). They already complied and removed the marketing language. Suing to reverse the ruling after you've conceded the reme…

Worth noting this is an anode-free design. Removing the anode matters more for commercialization than the energy density headline because it cuts material costs and simplifies manufacturing. 80% capac…

The "citizen developer" framing assumes AI flattens the skill distribution but the data says the opposite. Next-token prediction and RLHF typicality bias mean AI converges to the mean, so do…

People are actually building this stuff. Not much but I built an ML-powered RSS reader with a swipe interface that learns your preferences locally, no feed algorithm, no platform: https://p…

This essay pulls together a few threads from pieces I've written over the past months on my blog, where I dig into rabbit holes like GLP-1 economics among other things. If the pharma side interes…