Tech companies restructured safety committees.
Also illegal books and stapler wars.

The Office Reorganization That Fixed Nothing

OpenAI announced a corporate restructure, which is corporate speak for cleaning out the desk drawers of people who quit and replacing them with a new committee. The move follows the departure of key safety people, including co-founder Ilya Sutskever, who was apparently tired of the entire safety drama. OpenAI insists that its "Safety and Security Committee" will now advise the board, which sounds exactly like the old safety team except this time, the board has plausible deniability. The committee is essentially a highly-paid internal audit of itself.

The previous "Superalignment" team, whose entire job was to make sure the AI does not turn us into paperclips, was disbanded. It seems the management decided the best way to superalign the AI is to give the people building it a new job title and a fresh email distribution list. It is an act of benevolent corporate incompetence; they are trying very hard to look safe while actively prioritizing the roadmap. The rest of us can only hope the new committee is good at PowerPoint presentations.

Meta Used the Wrong Source Citation

A US District Judge indicated that Meta may have illegally used copyrighted books to train its large language model, Llama. The judge essentially questioned Meta's defense that using an author's entire life's work to teach a robot how to write is "fair use" because the output is transformative. If a junior developer copies a block of code off Stack Overflow and changes the variable names, it is plagiarism, and they are in the exact same kind of trouble.

The tech giant is facing a class-action lawsuit from several authors who feel that their intellectual property should not be used as free training data for a machine that competes with them. Meta is currently arguing that reading a book is the same as digitally ingesting millions of them, which is the kind of logic only a systems administrator who is very tired of fixing legal department printers can truly appreciate.

We Have Come Full Circle on Container Orchestration

A developer posted an argument for replacing Kubernetes with systemd for many common server applications. Kubernetes, the complex and highly-scalable darling of the cloud, is apparently too much for most workloads, which is like admitting you bought a combine harvester to mow a small suburban lawn. The entire article is a highly detailed, 400-point argument for simplifying your infrastructure back to the way it was five years ago.

The core observation is that many teams only use a fraction of Kubernetes' capabilities and spend a disproportionate amount of time managing the complexity they do not need. The proposal suggests that basic container management, logging, and monitoring can be handled by systemd, a utility that already exists on most Linux systems. We are now paying consultants to teach us how to stop using the complicated thing we paid consultants to install.

Briefs

• AWS Security: AWS's new security tool contained a security risk. It seems they installed a lock box but left the key taped underneath the hinge.
• The Death of Daydreaming: A cultural commentator wrote a piece on the death of internal thought. This suggests that the human brain's free trial has finally expired, and we must now pay for the premium subscription to think about what to have for dinner.
• Movie Tariffs: Former President Donald Trump announced 100% tariffs on foreign films. This is the government attempting a hard block on a high-bandwidth visual stream; the latency on popcorn prices will be astronomical.