OpenAI changed the model setting again.
Also tiny computers and employees trying to work offline

The Quarterly Corporate Clean-Up Just Deleted Your Favorite Feature

The tech giant OpenAI decided to pull the plug on its GPT-4o model for paying ChatGPT consumers without much in the way of a public memo or warning; it just vanished. This is the digital equivalent of IT decommissioning the only server that ran your favorite legacy spreadsheet application over a long weekend, then saying it was a "surprise roadmap pivot". It simply makes you wonder if the company's internal knowledge base is just a single sticky note on an unpaid intern's monitor.

It is a classic case of benevolent incompetence; the company likely felt the other models were technically "better" or more cost-efficient for the enterprise architecture, so they removed the model the users actually liked to keep the server racks tidy. The users are now left staring at the empty space on the dashboard where their beloved, slightly unhinged AI used to be, much like an employee returning to their desk to find their favorite ergonomic chair has been replaced with a wobbly plastic stool. The goal is always stability and cost savings, but the result is always chaos and resentment.

The Users Are Building a Server Room in Their Closets

The eternal struggle between the corporate cloud and the rebellious user base has reached its fever pitch; employees are now building their own AI workspaces entirely offline. This is what happens when the central IT department is too busy arguing over what version of the coffee machine firmware to deploy. Users are tired of their tools changing under the guise of an "update" and the constant surveillance required to run the latest large language model.

One user on a discussion forum pointedly asked how a behemoth like ChatGPT can serve 700 million users when a single individual cannot even run one instance of GPT-4 locally. The answer, of course, is scale, money, and a lot of duct tape; but the sentiment points to a deeper truth. The local-first movement, exemplified by the excitement around the Linear issue tracker, is a declaration of independence; it is the realization that sometimes the best IT solution is the one the IT department cannot touch. This quiet revolution is leading to users like Principal Engineer David Heinemeier Hansson praising the powerful modularity of the Framework Desktop because they just want a machine they can actually own and repair for once.

The Office Manager Made a Business Card That Just Works

Some of the best engineering is simply a spectacular waste of time, which brings us to the ultrathin business card designed by Nicholas L. Johnson. The card does not just contain a QR code for a LinkedIn profile; it runs a beautiful fluid simulation. Yes, a business card, a flimsy piece of paper you immediately lose in your wallet, now requires a GPU to render its core functionality.

This achievement is the ultimate power-move in a world obsessed with efficiency. While others are busy optimizing their cloud spend and reducing their data center footprint, Mr. Johnson is quietly delivering the world's most impressive, least-necessary piece of personal marketing collateral. It is a sign that innovation is not always about solving grand problems; sometimes, it is just about flexing a little too hard on a piece of cardboard.

Briefs

• LLM Intelligence Test: Someone tested a supposed pre-release version of GPT-5 by asking how many times the letter 'b' appears in 'blueberry'. It failed; the AI is still struggling with fourth-grade counting drills.
• The E-Paper Monitor: The Modos Paper Monitor is an open-hardware e-paper screen; it promises to be the perfect monitor for when you want your display to look exactly like the stack of TPS reports you have been avoiding all week.
• Tiny, Cheap Computing: A new RISC-V single-board computer is available for less than 40 euros; it is the ideal component for your personal, off-the-grid server closet rebellion.