이미지 생성 고도화, 개인 에이전트 확장, 수학 추론 비용과 에이전트 실행 구조

ChatGPT Images 2.5—faster, sharper, smarter, with better tools for creating whatever you can dream of. - Faster image generation to keep your ideas flowing - Improved fidelity for more natural, recognizable images - Consistent details across multiple edits - Comment-based

More exciting news from @OpenAI : GPT-Image-2.5-Sunburst is #1 and Flare is #2 across the Text-to-Image Arena, the Image Edit Arena, and the Multi-Image Edit Arena! Both releases are significant improvements over GPT-Image-2 (medium) in all 3 Arenas. Sunburst delivers the highest

signüll

@signulll

1일 전

muse has been a genuinely impressive product to play with. it has all of the functionality of the imessage agents in flight today but also has all of the ingredients that will expose it to hundreds of millions including a massive friend graph through insta, increasingly rich context from email & other services which you connect, & more importantly stuff like fb marketplace. marketplace in particular is an incredible distribution wedge. millions of normal people could encounter muse simply cuz an agent helps them find something, negotiate the price, & arrange pickup. pretty good execution here from fb.

Today we launched https:// Muse.ai, a personal AI agent, and published a deep dive on how we built safety into its system. An agent that gets to know you over time necessarily holds a lot of context about you. That's the source of its usefulness and the reason we built

For the Navier Stokes proof OpenAI used 130 billion output tokens or $6.50M with API pricing implying trillions to tens of trillions of input tokens for a total cost of slightly below $10M up to like $30M - $40M *assuming its the same scale as GPT-6 and just further RL'd

in my talk on how to build benchmarks i say you need to make your benchmark very challenging, but not too challenging, and i always say "for example the millennium prize problems could be a benchmark but it would be too hard it would take 100 years to saturate that" oops

Using more tokens doesn’t always mean better results. 👀 The real measure of AI coding efficiency is whether an agent has the context it needs to move work forward (without wasting time or resources). Here’s how we optimized GitHub Copilot to reduce unnecessary work while maintaining task quality.

Caspar

@caspar_br

1일 전

Agent auth is hard, but it shouldn't be! Your agent needs to act as someone: sometimes that's one shared identity, sometimes it's the person asking. To get either right you'll need an OAuth app for each service, a token store, refresh logic, and a consent flow Today we're fixing this with Managed Connections: define who your agent acts as in code, and the rest is handled for you: 𝚊𝚐𝚎𝚗𝚝 -> agent uses its own token 𝚞𝚜𝚎𝚛 -> agent uses the user's token, guiding them through the consent flow if they haven't connected yet With a single argument you avoid the whole OAuth dance. Available now, in 𝚖𝚊𝚗𝚊𝚐𝚎𝚍-𝚍𝚎𝚎𝚙𝚊𝚐𝚎𝚗𝚝𝚜 0.7

Here's a fun one with @preston_fu et al -- how should we reward RL policies in a way that scales to longer and more difficult tasks? Our answer lies in-between RL and imitation learning, and provides a simple way to assign dense per-token credit to long trajectories.

Runway is now in Adobe. Access the world's best models directly inside Premiere Pro and After Effects to generate, edit and upscale without leaving your timeline. Get started with the new Runway Plugins at the link below.