OpenAI loses big money testing code.
Also a Judge told the Social Media Guy to clean his room.

The AI Department Forgot to Check the Receipt

OpenAI, the firm famous for having an AI that sounds vaguely confident, reported half-year financials that are a study in Silicon Valley arithmetic; $4.3 billion in income is the good news. The bad news is a $13.5 billion net loss, which their accountants have categorized as an "infrastructure oopsie." The numbers suggest OpenAI is very good at selling something, but equally adept at setting a massive pile of money on fire to train it.

Chief Executive Officer Sam Altman likely sees this as an investment in the future, much like buying a server rack that is ten sizes too large for the current office; the rest of the industry sees it as the cost of attempting to reach Artificial General Intelligence using the world's most expensive utility bill. The consensus in the comment threads is that the valuation hinges entirely on the belief that a massive GPU farm will one day pay for itself; this is the core of the AI bubble, treating compute cycles like accrued interest instead of sunk cost.

Security Team Pre-emptively Patches the Future

Signal, the privacy-focused messaging application, has announced the implementation of the Signal Protocol and Post-Quantum Ratchets, or SPQR for short. This is what happens when the security team has too much time on their hands; they are preparing for the day a quantum computer, which is not yet a threat, manages to break their current encryption keys. The company views this as a necessary, proactive patch against a theoretical bad actor who lives in the year 2045.

It is essentially a cryptographic insurance policy, ensuring that the communications you are having right now about where to order pizza will remain private even when the laws of physics catch up with the server room. The general attitude is one of appreciative weariness; sysadmins understand the need to install a firewall on the toaster oven, even if it feels excessive.

Meta’s Recommendation System Gets a Performance Review

A Dutch judge, dealing with a Bits of Freedom lawsuit, ordered Meta to respect a user's choice of recommendation system. This is the judicial equivalent of telling a child to share the toys. The company must now give users an option to choose a non-personalized feed, which is probably a feed consisting only of pictures of old family pets and sponsored posts for industrial cleaning supplies.

In a move that perfectly illustrates the company's commitment to compliance, Meta also announced that it will listen in on AI conversations to personalize ads. The takeaway is that Mark Zuckerberg, CEO, must now follow one rule while simultaneously inventing a dozen new ways to know what color socks you are wearing; the system works as designed.

Briefs

• AI Finds Bugs: An AI assisted tool discovered potential issues in the curl library. The irony of using the thing that broke the web to fix the web is lost on no one.
• Amazon Drone Mishap: Two Amazon delivery drones crashed into a crane in Tolleson, AZ. They probably argued over right-of-way, like two interns fighting for the last good coffee pod.
• Red Hat Breach: Red Hat confirmed a security incident after hackers breached a GitLab instance. See, even the most serious IT departments forget to update the license on the guest Wi-Fi sometimes.