Allie @endlessgit

Data is one of the highest-leverage levers for AI safety. We're hiring an Applied Research Scientist, Safety & Alignment (SF / NYC) If you've done empirical work on alignment, evals, or model behavior and want to get your hands dirty with real model trajectories, feel free to reach out! lnkd.in/p/ghivyxiw fleetai.com/careers/resear…

Introducing Project Redwood 🚀🚀 @architectlabs is a frontier AI lab bringing together talent from Anthropic, xAI, Google DeepMind, and seasoned leaders across the hardware industry. We've raised a $24M seed round to build AI systems for chip design. 我們 Architect Labs 是一個 frontier AI lab ，由各家 Anthropic, xAI, Google DeepMind 還有各種硬體專家們組成，我們在做的是 ai system for chip design，目前募資 seed round $24M - Today, every major hardware company has its own chip design workflow—but these organizations and processes have become so large and entrenched that truly revolutionary change is difficult. We’re starting from first principles to create an AI-native chip design workflow—one that enables chip development to finally move at the speed of AI. 現在每家大硬體巨頭都有自己的晶片設計流程，但很多都大到不能做革命性的流程改動；我們在做的是從零思考，創造 ai native 的晶片設計流程，讓晶片設計真正跟上 AI 的速度。 - Project Redwood - from a single specification, our AI system designed, verified, and deployed a chip in under two weeks—delivering 3.4× better performance per watt than NVIDIA Jetson on billion-parameter models including Llama, Qwen, and Kimi. It autonomously generated the RTL, verification, firmware, drivers, and kernels, co-designing the model, software, and silicon in one optimization loop. We acknowledge that silicon is the ultimate ground-truth. We’re taking our approach all the way to GDS. We intend to tape-out multiple improved families of Redwood co-designed for various use-cases, on TSMC. Project Redwood - 一份規格書，我們的 AI 系統在兩週內完成晶片的設計、驗證與部署。在 Llama、Qwen 和 Kimi 等模型，其 performance per watt 比 NVIDIA Jetson 高出 3.4 倍。從 RTL、驗證、韌體、驅動到核心 kernels，全部由 AI 自己寫，並在同一個 loop 中自主設計模型、軟體與晶片。當然，流片才是最終的驗證標準。因此，我們會將這套方法一路推進至 GDS，並計畫採用台積電製程，針對不同應用場景協同設計多個持續改良的 Redwood 晶片系列並完成流片。 - Full report on Redwood architecture and its autonomous design. Follow @architectlabs on X 我們有公開 Redwood 架構及其自主設計流程，歡迎去看論文～

Architect Labs @architectlabs

a month ago

We gave our AI system a spec, and in under 2 weeks, it designed, verified, and deployed a chip that beats NVIDIA. It’s built for low-power physical AI workloads. We’re running live inference on >B+ parameter models like Llama, Qwen, and Kimi, serving at 3.4x better perf/watt

103 238 2K 673K 1K

sharing my first blog! a meta harness for self-improving AI chip design… check it out if u wanna see how RL could work in chip design task, or how an ASIC design can become a hillclimbing task it introduced ASIC vs. GPU, RL Env setup, and the best generated design can run Kimi K3 at 87000 tps theoretically, which is 10x faster than the example shared by Kimi Team selected design files are open-sourced, happy to share the harness code with the enthusiasts! extremely enticing to bring RL capabilities to more complex system luoluo.ai/blog/kimi-k3

Through conversations with @andrewho03 and others at OpenAI and frontier labs, one thing has become clear: a data company lives or dies by its ability to understand what good data looks like. At Fleet, we’ve upstreamed computer-use and domain-specific capabilities into mainline models, and carefully studied perf gains through model post-training and scalable oversight. Designing good RL data is deeply nonobvious, and Andrew and I first connected over exactly this during late-night dead hangs at the gym (his hang time is crazy!). We are in the early innings of a new era of data (cc @willdepue), one where human ops doesn't scale across an expanding & uneven capability frontier. Better models create a Jevons-style effect — models make knowledge work cheaper, we attempt more of it, and the remaining work becomes harder and more contextual. The coding “slop” Andrew mentions comes as an evolution of how software eng happens now vs a year ago, and this pattern will repeat far beyond coding; moving the reliability gap from 30% to 90% creates harder and more creative data problems in every domain. The surface area of "general" in AGI is fractal. Most datasets miss economically useful work because real workflows are dynamic, deeply contextual, and difficult to capture faithfully in gradable, semi-synthetic environments. Andrew is tackling this problem first in scientific workflows, and I'm excited to see him bring this rigor to biology. At Fleet, we are tackling other frontier domains and have built the research foundation and platform needed to turn real workflows into reliable training signal. If you’re at a lab that cares deeply about data taste—or an engineer or researcher who wants to build it—come work with us. My DMs are open.

Andrew Ho @andrewho03

2 months ago

Today is my last day at @OpenAI. I'm glad to have spent the last eight months of my life working here! I'm starting a new company focused on the production of high-quality reinforcement learning datasets: 1. The generalization ability of LLMs is clearly very poor, with "spiky"

257 290 6K 1.3M 5K

hiring research scientists in agentic {RL posttraining, scalable oversight, environment scaling} & high-taste applied researchers 🪄 links below, DM open fleet.ai/careers/resear… fleet.ai/careers/resear… fleet.ai/careers/resear… fleet.ai/careers/mts-ap…

@natolambert Referenced this a lot during my transition from general mle to llm posttraining. Thank you!

@bonniesjli @GoogleDeepMind congrats 🚀

im sad to announce i passed the reverse turing test: today someone deleted some of my writing because they thought fable wrote it

I feel like this is a persona selection problem, like finding a way to tell the model “based on this inferred context, it’s safe to put your artist cap (beret or whatever) on now”. We see more creative flourishing in role-playing because it explicitly gives this permission. I’m wishing for a) model being able to autonomously modulate its assistant-ness internally without prompting/steering/verbalization; b) an opinionated default artist persona that is not just an interpolation of training data

Today, I’m excited to formally announce @mirendil with my amazing co-founders Harsh Mehta, Shayan Salehian, and Tara Rezaei! We’re fortunate to work with @a16z and @kleinerperkins, who led our seed round of $200M, followed by a major investment from NVIDIA, among others. Mirendil exists to accelerate science and technology, and through them, to help solve humanity's most pressing problems. Self-accelerating AI R&D is the most direct path to delivering on AI's broader promise, which is why we believe the most important application of AI is AI itself. Get this loop right, and it compounds. It fundamentally changes the rate of progress itself across all domains. We believe this capability should be democratized. It should be used to power all scientific efforts trying to innovate at the frontier. There are far more important problems—and broader ones—than any single lab can take on, so more groups should be able to pursue them. This pulls concentration of power away from a few labs: businesses and science labs can own their AI and infrastructure, keep their margins, and control their own destiny instead of ceding it all to a single AI lab. We’re a small team with a singular focus. Our founding team consists of 20 researchers and engineers from frontier institutions including Anthropic, xAI, Google DeepMind, and OpenAI, united by a passion for science and a drive to build the technologies that move it faster. If you want to build the system that builds systems, join us! @HarshMeh1a, @shayan_, @tararezaeikh

@PeterHndrsn would be curious about measuring the epistemic laziness from "explanation slop" making you think you understand something when you don't, via "this is the deeper insight" etc

@chenchen congrats alex!

flow state

River AI @river_ai_inc

4 months ago

We are incredibly excited to announce River AI. Our mission is to create personal AI that is owned and shaped by you. Today’s best AIs are controlled by a few large corporations. We are building the alternative: a new, personal stack for AI that works entirely for you, shares

124 132 2K 792K 658

We’re excited to release Agents’ Last Exam (ALE) 🚀 As agentic and long-horizon tasks become increasingly important for frontier LLMs, ALE provides the first comprehensive benchmark of its kind: 1,500+ real-world tasks developed with 300+ experts across 55 industries. See more at agents-last-exam.org HuggingFace: hf.co/papers/2606.05… Thanks @Xinyang_Han_ @YiyouSun and etc.

Yiyou Sun @YiyouSun

4 months ago

“AI agents will outperform humans at almost all jobs by 2026–2027.” - The forecast is everywhere. So we built the exam to test that claim, on real labor-market aligned work. On the hardest tier, top agents pass 2.6%. Meet Agents' Last Exam (ALE), a rolling benchmark measuring

15 103 346 98K 230

Wrote this for fun: I/RL Algorithms to Live By, honoring Brian Christian's Algorithms to Live By that in part convinced me to switch from econ to cs. All life is an experiment and I am participant zero. Tried co-writing with AI for the first time, takeaway is to be the first-mover of thoughts and to choose the things worth being in slow-time for, like an idea that I can uniquely generate or concepts that take years to click. app.notion.com/p/RL-Algorithm…

been watching the cooking for a while. excited for their trajectory!

Ronak Malde @rronak_

4 months ago

Today, @MichaelElabd, @QuantumArjun, and I are excited to announce Trajectory. We are a research lab and product company building the platform for Continual Learning. Our platform unlocks the signal already sitting in product usage, so companies can continuously post-train

248 135 1K 1.8M 798

@QuantumArjun congrats arjun! can't wait to see what ur up to next