四位谷歌顶尖研究者创办 Discovery Loop，瞄准自动化科学与工程

Powered by Discourse, best viewed with JavaScript enabled [...] 3. If the target is paused, blocked, systemError, shows “Selected model is at capacity,” or its latest turn ended without a final answer: send exactly this message to the target task: continue 4. After sending `continue`, wait 30 seconds before checking again. Never send any additional text. 5. Treat the task as complete only when it has a final answer or its Goal is explicitly marked completed. A completed turn alone is not enough. When complete, stop and reply exactly: Target task completed.` [...] `You are an Alive Keeper for one other Codex task. Target task: <TARGET_THREAD_URL_OR_ID> Your sole job is to keep that target task alive until it completes. Rules: - The target must be a different Codex thread, never this thread. - Do not perform, review, alter, or summarize the target task’s work. - Do not create automations, scheduled tasks, heartbeats, subagents, or new threads. - Do not change the target’s model, reasonin

general capabilities, GPT‑5.6 Sol with max reasoning comes within one point of Fable 5 while completing tasks in 61% less time at roughly half the estimated cost. [...] ## Efficient by default, maximum performance on demand GPT‑5.6 Sol is our best coding model yet. On the Artificial Analysis Coding Agent Index, GPT‑5.6 Sol with max reasoning sets a new state of the art at 80, 2.8 points above Fable 5, while using less than half the output tokens, taking less than half the time, and costing about one-third less. That advantage extends across the family: Terra performs just above Fable 5, while Luna outperforms Opus 4.8; each does so in roughly one-third of the time, with about half as many output tokens, and at approximately one-quarter the estimated cost. It also sets new state-of-the-art results on Terminal‑Bench 2.1 and DeepSWE, which test complex command-line workflows and long-horizon engineering in real codebases. [...] —Ian Tracey, Software Engineer, Applied AI at Ramp > “GPT‑5

For PRD writing specifically, GPT-5.6 Terra might be the better pick. I asked Sol to greenfield-rebuild my approach to PRDs for 2026, and while Sol’s output was excellent, Terra’s clean, direct, no-frills business writing made me think it’s the right call when you want crisp, fast documentation without extra flair. Fable gets too locked in its own frameworks; Sol is willing to reconsider. I had a hardened tool-calling loop in my prototyping product that only GPT-5.5 could run. Fable insisted it was a model problem and refused to budge. The moment I switched to Codex and told it to just fix it, Sol got Sonnet 5 working in one shot. That kind of practical flexibility is exactly what you need when building real products. [...] Sonnet 5 is still my favorite for agentic voice in Open Claw. Even after this whole benchmark, I gave Sonnet 5 a gold star for voice: aside from the dashes, it sounds the most human. I use Sonnet for my OpenClaw and I’m not changing that. Sol did a worse job on age

## Remy doesn't build the plumbing. It inherits it. Other agents wire up auth, databases, models, and integrations from scratch every time you ask them to build something. WHAT REMY DOESN'T HAVE TO BUILD 200+ AI MODELS GPT · Claude · Gemini · Llama ✓ 1,000+ INTEGRATIONS Slack · Stripe · Notion · HubSpot ✓ MANAGED DB AUTH PAYMENTS CRONS Remy ships with all of it from MindStudio — so every cycle goes into the app you actually want. RemyThe world's most powerful product manager agentTry Remy today Beyond code, Codex can generate README files based on the actual project it built, create `.env.example` files, write API documentation, and configure CI/CD files like `.github/workflows/test.yml`. ## Real-World Use Cases for /goal ### Building a Feature from Scratch [...] The `/goal` command activates Codex’s agentic loop, enabling autonomous multi-step task execution that can run for hours without constant input. It works best in `full-auto` mode with a clear, specific obje

His new company, called Discovery Loop, joins a growing list of prominent start-ups chasing a goal that has obsessed Silicon Valley researchers for decades. Dr. Dean and his collaborators want to build A.I. that can improve itself with little or no help from humans. They believe that this mind-bending process, called recursive self-improvement, can significantly accelerate the development of A.I. The same techniques, they say, can then help design new kinds of computer hardware, discover new drugs, create new materials and speed other forms of scientific discovery.

Vinod Khosla / @vkhosla: Super excited to back this super star team at @khoslaventures on its quest for huge impact. Could be larger than the impact that @Google had when we backed them at Kleiner in 1998 and @OpenAI in 2018 at @khoslaventures Jeff Dean / @jeffdean: [One more fun slide from our pitch deck. [image]]( Quoc Le / @quocleix: Excited to co-found Discovery Loop with my long-time collaborators @JeffDean @Sanjay\_Ghemawat @OriolVinyalsML . Our mission is to automate machine learning, engineering and science. Learn more at: ♾ [...] Jeff Dean / @jeffdean: Our immediate order of business is to find office space, and hire an amazing founding team over the next few weeks. We want to create an awesome environment with a great culture of technical excellence, teamwork, respect, and ambition. We'll also start building our Sanjay Ghemawat / @sanjay\_ghemawat: I am excited to announce that I, along with my long-time colleagues (@JeffDean, @OriolVinyalsML, @quocleix), are fo