Picking up where last week left off, super intelligence clocked in for a full factory shift, agents were graded on real professional work, and the true cost of building it all was laid out in public.
Robots worked a full shift, not just a demo. On November 3, Xiaomi published an eight hour continuous video showing two humanoid robots from its CyberOne 02 line sorting automotive components at its Xiaomi Auto factory, working without interruption through what the company presented as a full production day.
Why it matters for how we live and build: a robot that lasts a whole shift changes the question from whether humanoids can move to whether they can be scheduled, managed and trusted like any other part of a working operation.
Nvidia gave robots and cars new worlds to practice in. On November 5, Nvidia released Cosmos 3, four open world models of 16 billion parameters that generate multi camera driving footage in varied weather, synchronized camera and LiDAR data, and video predicted from robot motions, under a licence that also allows commercial use.
Why it matters for how we live and build: machines learn faster and more safely in simulated worlds than on real roads and factory floors, so better world models quietly speed up every robot and vehicle built on top of them.
Anthropic laid out the scale, and the stakes, in public. In its amended S-1 filing covered on November 4, Anthropic disclosed revenue of $33.24 billion for the first three quarters of 2026 and purchase commitments of roughly $895.4 billion, and said in its risk factors that, by its internal estimates, it may soon reach the threshold in its Responsible Scaling Policy for significant AI driven acceleration of progress.
Why it matters for how we live and build: these are no longer startup numbers, and a company telling public investors that its own systems may soon speed up their own development is asking all of us to think about oversight at the same scale as spending.
Agents were graded on real work, and the grades were humbling. SemiAnalysis launched GDPx, a benchmark of 206 tasks drawn from professional work, and the best model, Claude Opus 5.5, scored only 12 percent, with Google Gemini 4 Carbon close behind at 10.9 percent.
Why it matters for how we live and build: long, messy professional work is still far from solved, which is good news for human judgment and a useful reminder to delegate in stages we can check, not all at once.
The open weight wave kept rolling in from China. Moonshot published the downloadable weights for Kimi K3.1 Pro, which it calls its most powerful model yet, in the three trillion parameter class and able to understand images, while Tencent released Hy4, a new open model that understands images and is strong at finance, web development, game development and office tasks, with offensive cybersecurity training data deliberately left out.
Why it matters for how we live and build: powerful open models that anyone can download and run on their own infrastructure keep lowering the cost of building with super intelligence and keep real choice alive for smaller teams.
Security agents started doing the unglamorous work at scale. OpenAI reported an autonomous pipeline that found, validated and generated patches for 948 Linux kernel vulnerabilities without human intervention, and said its new GPT-6 Cyber model, released this week for defensive security work, found 81 new vulnerabilities in Firefox.
Why it matters for how we live and build: finding and fixing flaws before attackers find them is some of the most valuable work super intelligence can do, and doing it at this scale changes what a small security team can realistically protect.
Agents clocked in at work, and in our money. Perplexity launched Teammates, continuously running workplace agents that are given goals rather than step by step instructions and can take over existing projects in an isolated environment, while Anthropic launched Claude Personal Finance, which lets subscribers link bank accounts that Claude can read but not move money from.
Why it matters for how we live and build: super intelligence is moving deeper into the tools we already use and into the most sensitive parts of daily life, so clear permissions, read only limits and the ability to check what an agent did are becoming everyday necessities, not expert concerns.
Publishers finally got one shared way to say no. On October 30, OpenAI, Google, Microsoft and Anthropic said they will back one shared robots.txt signal, the token __usage-ai-training, so a site can add a single entry to say its pages should not be used to train generative models, instead of maintaining separate rules for each company. Anthropic says it has already started implementing it, while the others plan support in the coming months and into 2027.
Why it matters for how we live and build: culture is the training ground of super intelligence, and a simple, shared way for creators and publishers to state their terms is a small piece of infrastructure that protects the work everything else is built on.








1 Comment