ao3.life was scraping fanfiction, packaging it into AI-generated audiobooks, and selling it — without asking anyone. Authors responded by burying it in takedown notices until it went dark. A clean, coordinated counter-strike that didn't require a lawyer or a regulator.
AI-generated addons are flooding WoW's ecosystem so fast that Warcraft Logs is locking down its API to stop the damage. Legitimate developers are watching their work get cloned and degraded in real time. The open plugin economy is becoming a liability.
The "weight-space programming" paper showed frontier models like GPT-6 Astra natively manipulating neural weights — and now Willison confirms Qwen 3.8 27B can do it locally. This is a capability threshold that matters: models editing models, no cloud required.
The All-In crew dissects Anthropic's IPO filing, which reveals $895.4 billion in commitments — a staggering number for a company still burning cash on frontier research. Google's simultaneous $500B market cap wipeout frames the moment perfectly: the AI trade is bifurcating between builders and incumbents.
Reducto's CEO argues the document layer — PDFs, scans, spreadsheets — remains the unsolved bottleneck blocking AI agents from doing real work in healthcare, legal, and finance. It's unglamorous infrastructure, which is exactly why it's a serious business opportunity.
Morgan Stanley is claiming a first-mover flag on tokenizing real-world assets, putting equities and Treasuries on a blockchain. With a Democratic sweep adding regulatory uncertainty to the AI trade, TradFi is quietly moving its chips toward crypto infrastructure anyway.
Mollick ran a live experiment on X's AI slop comment problem and found it trivially easy to filter — which makes the platform's inaction a choice, not a technical limitation. His framing cuts through: the slop isn't a hard problem, it's an incentive problem. Engagement is engagement, even when it's fake.
Amplifying a sharp observation: Revolut has joined a growing list of companies treating their AI intelligence stack as a core strategic asset — something to own, not rent. The message is blunt: if your intelligence layer runs on someone else's API, you don't control your competitive moat.
OpenAI's new LifeVal benchmark tests whether models can handle mundane real-world tasks — apartment hunting, grocery planning — rather than math olympiad problems. It's an admission that current benchmarks measure the wrong things, and that the gap between lab performance and actual usefulness is still embarrassingly wide.