Welcome, humans.
A developer on Reddit says they built a CAPTCHA that took Fable 5 10 minutes and 100K tokens to solve.
Incredible. We finally found a security system that creates the same emotional arc as making a human identify every traffic light in a grainy photo from 2009: confusion, bargaining, spiritual defeat.
The serious version: the old "AI can't solve this" goal posts keep moving. The new defense makes the task expensive, slow, and timeout-proof enough that even a frontier model gives up before the page does.
Can’t beat the bots? Run out the clock… or should we say run out the tokens?
Here’s what happened in AI today:
😼 OpenAI's first device is reportedly a screenless ChatGPT smart speaker with cameras, sensors, GPT-Live voice, and moving parts.
📰 Thinking Machines released Inkling, an open-weight multimodal model built around customization.
📰 Anthropic reportedly moved toward a possible IPO while backing Ode's $1.5B enterprise AI implementation bet.
🍪 OpenAI launched Codex Micro, a $230 keyboard for steering AI coding agents.
🎓 Today's AI Skill shows when to use ChatGPT scheduled tasks versus Workspace Agents.
...and a whole lot more here.
Hey: Want to reach 700,000+ AI-hungry readers? Advertise with us!
Love robots? We launched a robotics newsletter! Sign up here.
😼 OpenAI Wants ChatGPT to Move In
OpenAI's first hardware device is reportedly not a phone, glasses, or a laptop. It is shaping up as a portable, screenless ChatGPT speaker with cameras, sensors, GPT-Live voice, and movement.
According to Bloomberg, the device is meant to feel like a physical ChatGPT: rechargeable, movable around the home, able to see context, talk naturally, and control smart-home tasks.
So yes, OpenAI's first big consumer device may basically be: what if Alexa had frontier-model ambition and maybe a little too much eye contact?
Also... OpenAI released Codex Micro, a $230 keyboard for steering coding agents with command keys, status lights, a joystick, and a reasoning-effort dial. The speaker is the future product. The keyboard is the nearer-term test: put AI controls under your fingers instead of burying them in menus and chat boxes.
The new Codex keyboard…
Why this matters: The product is still expected around 2027, but it shows where OpenAI thinks the next interface fight goes:
rooms, not just chat windows;
voice and vision, not just typing;
physical companions, not just software assistants.
That is exciting if you want AI to handle messy household context. It is also the pitch that makes privacy people quietly grip the table. A screenless speaker with cameras, sensors, personalization, and lifelike movement asks users to trust OpenAI somewhere more intimate than a browser tab.
Our take: the legal subplot is already getting spicy. As we covered earlier, Apple is suing OpenAI for allegedly stealing hardware secrets. NBC News reported that Apple's lawyer mixed up two OpenAI employees while pressing the case.
That error does not erase Apple's allegations, but it suggests Apple may have sued without a clean picture of who did what. The case may depend on discovery, where Apple can demand internal emails, files, and hiring records it cannot see from outside OpenAI.
As for the tech... the keyboard and speaker point in the same direction: AI is leaving the app window. One gives agents physical controls; the other removes the screen and lets voice, sensors, and context become the interface. OpenAI is betting the next computing layer feels less like opening software and more like directing an always-present system.
Basically, OpenAI wants ChatGPT to become a thing you live with. The hard part is making that feel useful instead of, y’know… cursed.
FROM OUR PARTNERS
Spending too much on Claude API calls?
It’s not necessarily a given, it might be how your data layer is set up. We ran 56 live benchmark tests on a real enterprise query spanning Salesforce, Snowflake, and ServiceNow, and found that the right architecture cuts token usage by up to 97.6%, with per query cost dropping from $0.596 to $0.027 in some cases.
CData Connect AI makes that possible. Curious how?
🎓 AI Skill of the Day: Schedule the Task, Agent the Workflow
If a task repeats daily or weekly, do not automatically make it an agent. Ask first: does it need tools, or a reusable skill?
In OpenAI's short walkthrough, scheduled tasks handle repeat work like a weekday Chief of Staff brief, feedback triage, or a recurring Google Doc summary posted to Slack. Sweet spot: repeatable jobs using connected tools like Slack, email, calendar, Drive, Notion, or Docs.
The important wrinkle: scheduled tasks work for repeat tasks that use tools, but not Skills (reusable instruction sets or workflows attached to an agent). If the job depends on a Skill, make it a Workspace Agent so it can run in the cloud, keep the workflow attached, and be shared with your team.
Classify this recurring task:
Task:
[describe workflow]
Does it mainly need connected tools like Slack, email, calendar, Drive, Notion, or Google Docs?
If yes, make it a ChatGPT scheduled task.
Does it rely on a reusable Skill, custom workflow, or team-shared behavior?
If yes, make it a Workspace Agent.
Give me the setup, schedule, required tools/skills, and one failure mode.Favorite insight: schedule chores; agent workflows. Your future self deserves fewer recurring tabs. Want more tips? Check out our AI Skill of the Day Digest for July.
Have a specific skill you want to learn? Request it here.
🍪 Treats to Try
*Asterisk = from our partners (only the first one!). Advertise to 700K+ readers here!
Codex Micro gives you a $230 Work Louder keyboard for steering Codex agents with status keys, shortcuts, a joystick, and a reasoning-control dial.
Oak gives companies an identity-control plane for managing humans, apps, and AI agents - no pricing details.
Rime helps enterprises field customer calls with voice models trained on its conversational data - no pricing details.
Cell Cinema captures label-free 3D images of living cell states so biology teams can train models on cellular behavior - no pricing details.
RoboTTT brings test-time training to robot policies, stretching robot context to 8K timesteps without adding inference latency - research preview.
Grok Build is now open source, so you can inspect and customize xAI's coding-agent harness and CLI instead of treating the build loop like a black box.
Goodfire Silico runs interpretability and model-behavior experiments like replicating J-space on GLM 5.2, training reward models to reduce hallucinations, and explaining cancer-prediction features - private beta.
Click to watch on YouTube.
Goodfire cofounder and CEO Eric Ho joins Corey and Grant to explain what researchers are finding inside neural networks, from hidden concepts and confidence signals to curved mathematical structures that may help teams reduce hallucinations and design better models. Special shout out to SAS for sponsoring today’s video!
Watch and/or listen now: YouTube | Spotify | Apple Podcasts
📰 Around the Horn
Thinking Machines released Inkling, its first open-weight multimodal model, with controllable reasoning effort, 1M-token context, and a customization-first pitch.
OpenAI introduced GPT-Red, an automated red-teaming system that uses self-play to find attacks and harden frontier models.
Researchers reported experimental evidence of recursive self-improvement after an autoresearch agent spent eight days improving its own harness and beat a version humans had tuned for two years on held-out benchmarks.
Walden Robotics came out of stealth with Toyota backing, about $300M in funding, and a $1.1B valuation for industrial humanoid robots.
Apple Intelligence was registered with China's cyberspace regulator, clearing a path for Apple's on-device AI features on iPhones in China.
Google Research argued that diffusion-model creativity comes from learning smoothed score functions that interpolate along the data manifold instead of simply memorizing images.
Corey's read: Demis Hassabis raised the question of who should test the most powerful AI systems, as frontier labs build models that may outpace conventional evaluations and outside oversight.
FROM OUR PARTNERS
Startups are leveling up creative applications by embedding generative media models directly into intelligent workflows. Question is, which models work best for what tasks? Find out in the new Google for Startups technical guide on generative media.
🧩 Thursday Trivia
You know the drill: one is AI, and one is real. Vote below!
Image A
Image B
Which is AI, and which is real?Which is AI, and which is real? The answer is below, but place your vote to see how your guess everyone else (no cheating now!) |
A Cat’s Commentary
Bluff away, me hearties yo ho
Trivia answer: A is AI and B is real (from Poltergeist 3, a very haunted movie set).
 | That’s all for now.
|
Going for an anime aesthetic this month!
