Tech News

RL

Top post: @ryolu_ “wherever ideas hit. desks and laptops optional.”

AA

Top comment: @AIatMeta “To help accelerate neuroscience breakthroughs, we're releasing the full training code for Brain2Qwerty v1 and v2, and our partner, @bcbl_, is releasing the v1 dataset. Learn more and explore the artifacts here: https://go.meta.me/42ed9c”

KA

Top post: @yacineMTB “For every software provider, your biggest competitor is the customer themselves -- building it for themselves and thereby reducing reliance on you. Instead consider making a net positive business that helps your customer, instead of trying to rent seek forever”

SU

Top post: @Suhail “Excellent.”

RP

Top post: @rohanpaul_ai “Aravind Srinivas just explained why China’s open-source AI may become more powerful than ever. And why Anthropic has lobbied very hard for export control. "the only reason why there is even a 12-month gap between open source and frontier models is export controls. But there is a chance that, because of that, they now get really good at the physical layer. One advantage they (China) have is that they can actually build data centers a lot faster. Power is not a problem. Permits are not a problem. People are not a problem. Labor is not a problem. Expertise is not a problem. And so, by forcing them to go out there and build all this, you are converting them into a far more potent competitor." --- From "20VC with Harry Stebbings" YouTube channel (@HarryStebbings ), link in comment”

CH

Top post: @kimmonismus “"Meta, which has been among the biggest customers of Claude Code, set up the applied AI engineering team earlier this year and has tasked it with improving Meta’s own coding assistant, MetaCode. A big part of that entails building high-quality datasets and programming challenges that engineers use to train and test its models for coding work. While Meta permits the team to use outside AI tools for some purposes, it requires engineers to design those challenges themselves and to rely on their own technical expertise rather than AI-generated concepts."”

GT

Top post: @garrytan “Build power and datacenters”

SW

Top post: @ScottWu46 “Model routing is a hot topic but there are two challenges to doing routing for coding agents: 1) Even if different models can pass the task, there are subtle differences in behavior & style that mean they aren't perfectly interchangeable. 2) The initial agent prompt isn't enough to know the difficulty of the task. "Fix xyz bug" could be a one-line edge case or could require rearchitecting your entire product; you can't know until you've actually investigated the code. How do you solve these problems? Well, you need evals that account for style and behavior, not just pass/fail. And you need the agent to be able to dynamically update and re-route. We built Devin Fusion with both of these points in mind and found that it reduces costs by 30-40% while still maintaining the frontier intelligence "smell":”

T(

Top post: @teortaxesTex “Damn, that's right. So: "≈48B" (thanks to N-Gram embedding, variable) active, 35T tokens. V4-tier, ≈8e24? Would be the biggest Chinese pretraining on domestic hardware. Some strange "Superpods": > "our accelerators" > "up to 48 machines each" > below 80 GB HBM per accelerator > "The device offers limited HBM bandwidth but a relatively large L2 cache" > "built-in 200 Gbps network adapter within the accelerator" what is this stuff Was the first truly big Chinese model trained on domestic compute also using some extremely obscure piece of hardware?”

SP

Top comment: @SPACEXCOIN_CTO_ “@elonmusk What I think of when you talk about engines is the rocket engines at @SpaceX and that's really amazing because not everyone understands them. But are there other engines you're referring to that we don't even understand?”

JG

Top post: @jgebbia “Introducing Rampart, an open source model to protect the privacy of citizens’ personal information, built by the talented software team at @ndstudio. Watch the recorded live stream below with @as400495 @taigrr 🇺🇸”

SC

Top post: @MilksandMatcha “When OAI first launched deep research, I paid $200 for the subscription (which at the time was a CRAZY subscription price) and the first thing I did was ask it to deep research me. It's both terrifying, exciting, and unbelievably difficult to index humans. Pumped to see the team launch @CLODOAI @sixtensor @vikchuppala”

DP

Top post: @DimitrisPapail “@jm_alexia @HessianFree Welcome!!!”

CH

Top post: @kimmonismus “Really excited for the next version of Claude Code. Keep it coming Boris!”

AP

Top post: @APompliano “@chamath Congratulations, Chamath”

AV

Top post: @ashleevance “On this Iridium deal and RKLB's next launch The distance that SpaceX and Rocket Lab are putting between themselves and rivals is starting to look insurmountable. Rocket Lab still needs Neutron to fly and fly often and this is no small feat. But these two companies have put together so many pieces and remain the only ones that can launch faster than everyone else after ALL these years. They have just figured things out that others can't. Rocket Lab is so valuable now that the math gets hard. But it just feels more and more like we head toward some world where it's merged with Blue Origin”

TI

Top post: @thsottiaux “Do you think it has a reset button?”

DW

Top post: @deanwball “Could not come at a better time. I look forward to cheering my FAI friends on as they embark on this critical endeavor that FAI, uniquely, is suited to undertake.”

AC

Top post: @AndrewCurran_ “Finally a date for the new DeepSeek. The Whale arrives in about two weeks.”

RO

Top post: @tszzl “@gabriel1 yes but the question is why anthropic has had almost no dissent among the ranks”