But that's not the only thing surfacing. The system files are a goldmine for anyone tracking the roadmap. We're seeing mentions of the iPhone 18 series (V63, V64, V67), a foldable iPhone Ultra (V68), and even an "iPhone Air 2" (V62). It looks like the Pro models and the foldable might hit this autumn, while the standard 18 and the Air 2 are likely spring releases.
Moving over to the LLM side, OpenAI is playing it safe—or cautious. They actually d reinforcement learning (RL) for their frontier models for two weeks to tighten up their "safety gates." Apparently, they had some security hiccups with Hugging Face evaluations, and their upcoming Astra model might be hitting "critical cybersecurity capability" thresholds. To prevent a runaway agent or a data breach, they've implemented stricter code and network isolation. They are essentially running a hands-on guide for how to scale AI safely: test small, verify alignment, then ramp up.
Meanwhile, Anthropic is printing money. Reports suggest their annualized revenue run rate has cleared $65 billion. Most of that is coming from Claude's enterprise adoption and developer API usage. While they aren't officially IPO-ing tomorrow, the infrastructure for a public offering is clearly being laid.
On the hardware front, Xiaomi is teasing the next generation of its "Xuanjie" chips. Lu Weibing mentioned that the O1 has already shipped in over a million units across three devices. However, the financial side is a bit messier, with adjusted net profit dropping over 42% in Q2, partly because their EV and AI innovation arm is still burning cash despite delivery growth.
A few other quick hits:
Relay, the AI automation startup, is shutting down with parts of the team migrating to Google Chrome.** Enterprise WeChat**is opening up its CLI andMCP, allowing AI agents to trigger about 10 different office productivity capabilities.Unitree Robotics officially went public on the STAR Market with an IPO price of 150.80 RMB.
It feels like we're hitting a phase where "AI" is moving from just a chat box to actual eyes (AirPods) and limbs (Unitree), while the big labs are realizing that the faster the model gets, the harder it is to keep it in the box.
Leopold Aschenbrenner lost 30 billion dollars by betting too 23m ago
ChatGPT is becoming a primary emotional outlet for people who 1h ago
[Does anyone actually care about llms. 7h ago](/en/news/6847/)
[OpenAI just killed off its catastrophic risk team 16h ago](/en/news/6786/)
Joining an AI startup while my social circle actively despises 1d ago
AI safety regulations are becoming a convenient shield for 1d ago
Next Leopold Aschenbrenner lost 30 billion dollars by betting too →