VeryCodedly This Week: Paramount's $110B Takeover, OpenAI Watermarks EU Text, Margaret Hamilton Dies + More

Paramount completed its $110 billion acquisition of Warner Bros. Discovery, consolidating more than 200 million streaming subscriptions and nearly $70 billion in revenue under one roof as the company targets at least $6 billion in annual savings within three years. Separately, OpenAI is adding invisible watermarks to ChatGPT and Codex output across the EU to comply with the AI Act's machine-readable text requirement, using a system called textGrain, following Anthropic's similar move in August. Margaret Hamilton, who led the MIT team that wrote the Apollo mission software and popularized the term "software engineering," died on September 30 at age 90.

It’s a new week, and shockingly, more stuff managed to happen. Here’s what we picked for you. Skydance is live. Paramount completed its $110 billion acquisition debt included of Warner Bros. Discovery, so a lot of your favorite shows and movies, from Harry Potter to Pimp My Ride, are now under one roof. More than 200 million streaming subscriptions and nearly $70 billion in revenue. The company’s eyeing at least $6 billion in annual savings within three years and plans to merge the streaming services into one product eventually. This could mean fewer choices, higher prices, or both. Are you in the EU? Do you use ChatGPT to write emails, reports, school assignments, etc? Yes? No? Keep reading. OpenAI is adding invisible watermarks to ChatGPT and Codex output across the EU to comply with the AI Act, which requires gen-AI companies to make text output machine-readable. The system is called textGrain. The model chooses words according to a secret key, so across long enough text, the choices form a pattern only OpenAI’s detector can recognize. The watermark can help a detector identify text generated by OpenAI, but isn’t guaranteed to detect every passage, especially short or heavily edited text. Anthropic started this in August, so OpenAI isn’t the first. There’s also no universal AI-text checker, and the watermark can’t identify the user. Access to its detector is restricted to approved researchers and expert organizations. For now. This is the AI agent startup whose $2 billion acquisition by Meta was blocked by Chinese regulators. The round was led by Boyu Capital and IDG Capital, with existing investors Tencent, HSG, and ZhenFund making modest contributions to raise $500 million. Short timeline: Meta announced the acquisition in December, Chinese authorities intervened in April reportedly over concerns that scarce AI talent and tech would flow overseas. The big funders bought the shares back from Meta. Meta’s Muse and OpenAI’s Dots are now major competitors, but Manus raised half a billion dollars on its own terms. Not too shabby. AI features have been there for a while, but now they’ve reorganized the interface around agents. In the Agents window you can now run multiple agent sessions in a grid, drag them into horizontal or vertical splits, resize panes, and maximize to focus. Copilot sandboxing is now available across Windows, macOS, and Linux, limiting agents’ access to files and network resources. External sessions from Copilot CLI and the Copilot app are auto-detected so sessions follow you, no reload. A new “Chat: Open Worktree Cleanup” command lists inactive agent session worktrees, shows their size, and lets you delete them. It’s a peek into where Microsoft thinks software development is headed. On purpose, for security. The LZ77 dictionary coder is now disabled because shared state across channels could leak plaintext under certain conditions. Compression still works, just less effectively. Secondly, it rejects $ and \ characters in command-line usernames. If a username comes from an untrusted source like a CI job, or an internal tool, those characters can end up interpreted by the shell. Usernames set in config files are fine though. Workflows will be affected. But sometimes the secure thing is just to stop doing the thing. She led the MIT team that wrote the Apollo mission software and is credited with popularizing the term “software engineering”. Before that, people saw it as typing. Crazy. Hamilton managed the Software Engineering Division of the MIT Instrumentation Laboratory, leading a team of 100 engineers that designed the onboard flight software for the Apollo missions. Her software helped Apollo’s onboard computer recover from unexpected errors, including during the Apollo 11 landing. Turns out the Moon was not the place to discover your software has issues. She died on September 30 at age 90. Alright folks, that’s it for this week. See you next week.