OpenAI launches a robotics division, steipete trains Codex as a QA engineer, dotey on wielding multiple models — June 1
12 qualifying posts from 5 authors on June 1. Sam Altman formally names OpenAI Robotics with a live hiring push (12,645L). Steipete's Codex-as-QA-engineer setup draws attention (1,541L). Dotey posts two practical tips: model selection strategy and /goal as a planning primitive. Turingou on the Codex 0.6% stat and AI creation joy from Izu. Sophia with four art objects. Nyarime on free Minimax M3 API.
Yesterday's highlights from your network: Sam Altman formally opened OpenAI Robotics with a full hiring push, steipete's Codex-as-QA-engineer setup is getting attention, dotey posted two sharp tips on squeezing more out of Claude Design and the
/goal command, and Sophia's history feed brought four objects worth slowing down for.Sam Altman announces OpenAI Robotics — for real this time
After four days of near-silence, Altman came back at 16:07 UTC with the highest-engagement post of the window by a wide margin.
"OpenAI Robotics is hiring, looking for exceptional full-stack hardware, ops, systems, and ML engineers to help us program and manufacture robots that are useful for society... Our world simulation research program, led by Aditya Ramesh, has evolved over the past year into OpenAI Robotics." 1
This is the first time "OpenAI Robotics" has appeared as a named division with its own recruiting channel (
robotics-recruiting@openai.com). Near-term focus is on supporting skilled workers building infrastructure; long-term goal is a personal robot for everyone. The post hit 12,645 likes and 2.6 million views in under 24 hours.正在加载内容卡片…
Steipete: Codex runs QA so he doesn't have to
The most technical post in steipete's recent build log — and one of the sharper agentic demos circulating right now:
"Been teaching codex to be my QA assistant. For every commit it creates a user-test scenario and uses webVNC (crabbox), computer/browser use (peekaboo/mcporter) to test OpenClaw like a user/QA person would. This runs in the background and opens PRs with fixes." 2
The setup pairs Codex with browser / computer use tools to simulate real user interaction after each commit — closing the loop between "model writes code" and "model checks whether the code actually works for a user." 1,541 likes and 81K views.
正在加载内容卡片…
Dotey: two tips that hold up in practice
On model selection: dotey argued against picking one model and sticking with it. His framing: use them the way a savvy contractor picks tools — discover each one's actual strengths, then combine. Specific call-out: Opus 4.8 does better than GPT-5.5 on UI design and implementation, even if it falls short on pure writing tasks. Recommended workflow is Claude Design → hand the output to both models separately and compare the delta. 3
On
/goal as a planning primitive: his /goal workflow is essentially treating a Codex session as a project with phases rather than a prompt. Two uses he highlighted: (1) give it a checklist of subtasks in a JSON file, let it tick items off in batches; (2) write a multi-phase design doc with acceptance criteria per phase, then execute phase by phase with a commit after each turn. 4 329 likes.He also posted a tip on using Claude Design with imported design systems — Adobe Spectrum 2 as a specific recommendation — noting that a shared token system gives AI-generated designs more coherent visual consistency. 5 355 likes.
正在加载内容卡片…
Turingou: two dispatches from Izu
Turingou is traveling in Izu. Two posts made the threshold.
The first, paraphrased: he's been using Codex on mobile through a VAS connection, which pushed him toward mobile-first design across his products. He wrote a hook that syncs any updated web app to a new Expo build — cutting mobile development time in half, at the cost of a backlog of apps waiting for App Store review. 6
The second is a single observation that landed harder: Codex's user base is 0.6% of ChatGPT's. If you've burned through a billion tokens in Codex, you're a statistical outlier. His line: "Everything is just getting started." 7 277 likes.
And a quieter one: "The moments in life when you can fully enjoy the pleasure of creation are not long. I've already experienced enough happiness. Now, being able to create (almost) anything with AI makes me feel like the world might really be a simulation." 123 likes. 8
Sophia: four objects
Four posts from @SophiaFioren that hit the threshold during the window:
- Medieval pathway in Terni, Italy, 13th century — 295 likes. 9
- Huma bird figurine from Tipu Sultan's throne canopy, 1792 — 256 likes. 10
- Box-form ornament with granulation and filigree, Greece, 500 BCE — 157 likes. 11
- Armchair, England, 1755 — 131 likes. 12
正在加载内容卡片…
Also noted
Minimax M3 API: @realNyarime posted that free Minimax M3 API access is available by scanning a QR code — 437 likes, 97K views in the window. 13 Worth knowing if you're looking for a capable model to test without cost.
围绕这条内容继续补充观点或上下文。