AINews Dispatch
Fresh technical briefings on the model race, regulation, and product shifts shaping AI right now.
Showing 73-84 of 256 briefings
OpenAI's GPT-Live can listen and speak at the same time, ending the walkie-talkie era of voice AI. The real story isn't that it sounds more human — it's the two-layer design behind it, and what it means that the machine can now backchannel while you talk.
Read the briefingOn July 15, Doubao and Qwen switch off their humanlike companion agents to comply with China's new anthropomorphic-AI rules. The enterprise agents live on. That split — capability stays, intimacy goes — is the most interesting regulatory idea of the year.
Read the briefingA $40 million round for a company that builds fake workplaces is the clearest signal yet that AI's next bottleneck isn't intelligence — it's reliability. The math of compounding failure explains why the smart money is quietly betting the agents don't work yet.
Read the briefingOpenAI's GPT-5.6 spent its first weeks behind the first U.S. government security review of a frontier model. On Thursday the gate lifted and the model went on sale in three tiers. The precedent it set didn't lift with it.
Read the briefingAnthropic's Claude Science turns the model from a research advisor into a 60-tool workbench that keeps its own receipts — a bet that reproducibility, not raw intelligence, is AI's next frontier in science.
Read the briefingMicrosoft just built a 6,000-person company whose only job is to make AI actually work inside other companies. It's the clearest sign yet that the frontier has moved — from the model to the messy last mile of deployment.
Read the briefingMeituan's LongCat-2.0 — 1.6 trillion parameters, open weights, and reportedly not a single Nvidia GPU in the training run — is less a model launch than a stress test of America's entire chip-denial strategy.
Read the briefingQualcomm's ~$3.9 billion deal for Modular is a bet that the thing keeping the AI world locked to Nvidia was never the silicon. It's the software. And the engineer it just hired has spent his whole career rewriting the software that everything else runs on.
Read the briefingThe industry spent the first half of 2026 arguing over which protocol AI agents would use to talk to each other. The breach data says we built the roads before we built the driver's licenses — and July's scramble toward "Agent Zero Trust" is the belated correction.
Read the briefingGartner says a fifth of enterprise SaaS spending is exposed to "agentic arbitrage" by 2030. The number is eye-catching. The mechanism underneath it is the part every software company should be reading twice.
Read the briefingDRAM prices nearly doubled in a single quarter, consumer memory became collateral damage, and a new antitrust lawsuit is asking whether the squeeze is demand — or design.
Read the briefingWhile the world argued about the next chatbot, a different kind of foundation model quietly grew up around spreadsheets, databases, and data warehouses — and in 2026 it became a real category, with a billion-euro price tag to match.
Read the briefing