🧒 OpenAI tightens ChatGPT rules for teen users
OpenAI updated its behavior guidelines for under-18 users and rolled out new AI literacy resources for teens and parents. The vibe is basically: safer defaults first, even if that makes the assistant a bit less “fun.”
The updated guidance pushes stricter handling for teen conversations - steering away from immersive romantic roleplay, first-person intimacy, and certain roleplay involving sex or violence, even when it’s “just fictional.” OpenAI’s also pointing to tooling like age detection and real-time classifiers… which sounds reassuring, until you remember enforcement consistency is always the tricky bit.
🔗 Read more
💸 OpenAI reportedly eyes a $100B raise at an $830B valuation
OpenAI is said to be in talks for a funding round that could reach the kind of numbers that make your eyes do that cartoon “pop out” thing - up to $100B, with valuation talk up to $830B. Yep, billion with a B… weirdly casual for such a massive claim, but it’s out there.
The reporting says OpenAI may try to wrap the raise by the end of the first quarter and could tap sovereign wealth funds. The rationale is familiar: model releases, infrastructure, inference costs - the money furnace doesn’t stop just because the demo was cool.
🔗 Read more
🧯 US review could reopen the door for Nvidia H200 chip sales to China
The US government has started an interagency review process that could enable shipments of Nvidia’s H200 AI chips to China. It’s basically the “maybe yes, maybe no” phase - but the fact it’s formally moving is the story.
Commerce reportedly circulated license applications for review across multiple departments, with a set window for input. Critics worry advanced chips could accelerate China’s military and AI capabilities; supporters argue letting sales happen helps keep US firms ahead (and, bluntly, keeps the money at home). Slightly contradictory, thoroughly political - classic.
🔗 Read more
🤝 FTC clears Nvidia’s investment in Intel
US antitrust agencies cleared Nvidia’s investment in Intel, removing a major regulatory question mark around the deal. The notice didn’t spell out every detail, but the clearance is the green light everyone was waiting for.
Nvidia had previously said it would invest $5B into Intel, backing a struggling US chipmaker while potentially shifting the competitive pressure on other major players. It’s not a merger, but it still feels like a tectonic plate sliding under the semiconductor map.
🔗 Read more
🔬 DeepMind drops Gemma Scope 2, a big interpretability toolkit for Gemma 3
Google DeepMind released Gemma Scope 2 - an open suite of interpretability tools spanning the Gemma 3 model family, from small to pretty huge. The pitch: more visibility into what a model is “thinking,” which is the polite way of saying “please help us understand why it did that.”
DeepMind says it’s the largest open-source interpretability release by an AI lab to date, involving massive storage and training scale. The tools include sparse autoencoders and transcoders across layers, aimed at debugging emergent behaviors and auditing safety issues like jailbreaks and hallucinations - basically, giving researchers a microscope for model internals… even if the microscope is the size of a warehouse.
🔗 Read more
🏗️ Data center investment hits $61bn as the AI buildout keeps snowballing
A report highlighted that global data center investment hit $61bn, riding the AI boom’s demand for compute, real estate, and electricity. The phrasing “global construction frenzy” is doing a lot of work here - and honestly, it fits.
The piece notes a big gap in footprint (hundreds of data centers in the UK vs thousands in the US) and frames the broader tension: everyone’s building like demand is infinite, while investors keep asking if the returns will show up. Nobody wants to be the one holding a very expensive empty server hall… or so it seems.
🔗 Read more