Cybercab 설계, Astra 배포, 장문 추론 최적화와 에이전트 인프라

The Tesla Cybercab doesn't have ANY brake fluid. It is the first production vehicle to have a brake-by-wire system. Electronic actuators are responsible for the clamping pressure, rather than hydraulic fluid that you see in traditional vehicles. This system improves efficiency since the brake pads can fully disengage. They can go entirely off or on. It also enables Tesla to fine-tune the braking to be as smooth as possible. PLUS, when the car is built with the unboxed process, Tesla doesn't need to run any brake lines, everything is seamless.

Elon Musk

@elonmusk

10시간 전

The Cybercab motor uses no rare earth metals, but maintains the same range! This was extremely hard to achieve.

Sam Altman

@sama

9시간 전

We are also excited!

Good old days at OpenAI in 2016: an agent stares at screen pixels, moves a mouse, and books a flight on United. We called it World of Bits, inside OpenAI Universe. 10 yrs later, Astra is reincarnated in the same universe. Even the naming is astronomically correct 😆 Universe was

Great tips on working with reasoning models. Normally you would append what the model figured out after the document and ask again. It turns out that where you put the reasoning trace changes long-context accuracy by 50 points. Transformers process causally, so a task state discovered late cannot guide the reading that already happened. For causal state update processors, providing the condition first can require exponentially less memory in the worst case than providing it last. Trace as State puts the collected reasoning trace before the long-context block on a fresh pass, so information derived earlier guides the rereading. The matched control, Trace Append, uses the identical trace after the context. On GraphWalks Parents, DeepSeek V4 Pro Preview goes from 29.2% on the initial pass and 43.0% with Trace Append to 81.8% with Trace as State. GLM-5.2 goes from 66.4% and 83.2% to 100.0%. Trace as State wins in 26 of 27 reported combinations of model, task and metric, with no architecture change required. Paper: https:// arxiv.org/abs/2609.02702 Chat with Paper: https:// academy.dair.ai/papers/trace-a s-state-reasoning-traces-as-conditional-states-for-long-context-transform-2609.02702 …

Keunhong Park

@KeunhongP

9시간 전

When we say Atlas has pixel-perfect camera control, we mean it. Atlas precisely follows input camera parameters, including non-planar projections such as the Brown-Conrady distortion model and the Kannala-Brandt fish-eye model. @BhamidipatiPan1 invented a novel method of camera conditioning and it works beautifully. 🧵 [1/N]

“HarnessDev: Can LLMs Create and Evolve Their Own Agent Harness?” Most agent benchmarks fix the harness and only evaluate the model. So this paper explores whether LLMs can build and improve the harness itself, including the execution loop, tools, context, recovery, and verification. They showed that they can already create competitive harnesses from scratch, but self-improvement is still brittle, with gains often failing to generalize across unseen tasks or different executor models. https:// alphaxiv.org/abs/2609.01437

Guillermo Rauch

@rauchg

10시간 전

“Feedback is a gift” has always been a favorite mantra of mine and a strongly-held belief. Now it’s fact. Each piece of feedback is someone gifting you a prompt for you to give your agents to improve your product ☺️ Thankful for everyone who takes time (or tokens) to critique our products, write to us or forward your agents’ transcripts.

QwenWork hits 30M users just one month into beta in China, with enterprise accounts making up over half! With 120 updates and 1,000+ new features in just 30 days, QwenWork stands as one of the fastest-growing workplace AI products globally, designed to empower users across industries seamlessly.

Wan3.0 is now generally available on Alibaba Cloud. • 30-sec generation • 1-click doc-to-video • Production-ready Check the detailed guide here: https:// click.alibabacloud.com/m/20000003076/ #AlibabaCloud #Wan #GenAI #AI

Qwen Developers

@QwenDevs

11시간 전

E-Commerce Bench is a small attempt to evaluate models in a specific business setting. hope it can offer a useful reference for improving model performance on specialized business tasks.

Matei Zaharia

@matei_zaharia

8시간 전

Check out the Lakebase VLDB paper for tons of detail on why and how Neon and Lakebase are built! We think this type of highly elastic architecture over commodity lake storage (S3) is going to be used for more and more infra as software development speeds up and agents do more of it.