AI broke the production build again.
Also Tesla hit a pothole and the data center needs 10GW.

The Intern Wrote Production Code; It Used a Bot

A security report on HackerOne points out the fundamental issue with treating Generative AI as a competent junior developer. The core problem, according to the report, is that the original developers of the system clearly did not understand what the AI actually produced; they just deployed it. This is the equivalent of a middle manager handing you a five-page printout from a vendor, saying "just implement this," and then being shocked when it turns out the vendor's code was written in undocumented Fortran.

The resultant security vulnerability was a near-total data breach, but we are treating it as an "oopsie" in the same way we treat a spreadsheet error that over-allocated the Q3 travel budget. The industry appears content to outsource core functions to a large language model and then shrug when the model provides code that is syntactically correct but fundamentally reckless. Apparently, if you automate the process of not understanding your own stack, you also automate the process of creating exploitable flaws.

OpenAI and Nvidia Reserve the Sun

OpenAI and Nvidia, two companies famous for using vast amounts of electricity, have decided that they are going to use even more vast amounts of electricity. The two companies have announced a partnership to deploy 10 gigawatts of Nvidia systems. For context, ten gigawatts is roughly the output of ten nuclear power plants; or enough to power several million homes; or enough to run one thousand interns on electric scooters for a decade.

This is not a partnership, it is a land-grab for electrons. OpenAI Chief Executive Officer Sam Altman and Nvidia Chief Executive Officer Jensen Huang have essentially agreed to build a massive shared utility closet for their respective pet projects. This deal assures everyone that the price of compute will continue to rise while the price of actual, useful human labor will continue to drop, which sounds like an excellent strategy for everyone involved, especially the people who own the servers.

Tesla Full Self-Driving Project Trip Fails Before You Can Say 'Disruption'

In what can only be described as a perfect microcosm of tech ambition meeting asphalt, a group of Tesla influencers attempted a coast-to-coast FSD journey but failed before they even reached the first town. The car, allegedly in Full Self-Driving mode, hit some debris in the road less than 60 miles from its starting point, resulting in a damaged tire and a prompt end to the whole affair.

It appears that the 'Full' in Full Self-Driving does not account for the concept of physical objects that are not already neatly categorized in a vast dataset. The road, as it turns out, contains unexpected variables, such as "a thing that fell off a truck" or "reality." Tesla, the company, is just trying very hard to prove its technology works and accidentally proved that the traditional tire-changing industry is, in fact, future-proof.

Briefs

• New AI Model: Alibaba's Qwen group has released Qwen3-Omni, a 'Native Omni AI model' for text, image, and video. It is a new tool to generate more workslop, which according to a Harvard Business Review article is destroying productivity.
• Local-First Applications: A thoughtful piece asks why local-first applications have not become popular. The answer is obvious: you cannot sell a subscription or harvest user data if the user actually owns the data, which is a major design flaw in the business model.
• Retail Surveillance: Kmart, the retail giant, was found to have been unlawfully using facial recognition to tackle refund fraud. This is a classic case of a company trying to save five dollars by investing a million in a creepy, illegal solution.