LLMs, robotics, ML infrastructure, and AI applications.

Anthropic’s best agent finished 24% of real office tasks. OpenAI’s managed 8.6%. Now a $65M bet says enterprises will pay for the infrastructure layer anyway.











DeepMind has 7,000 papers, three Nobel citations, and a parent that prints cash. Everyone else is still scrambling for the same thing Hassabis bought in 2014.
OpenAI killed Sora burning up to 15M/day at peak against ~2M lifetime revenue. Disney got one hours notice on a 1B deal. And the 852B valuation rests partly on contingent capital that may never arrive.
An AI that answers your questions—or one that sells you things. That line just got blurry.
Engine Cinema doesn't record light — it interprets it. OpenAI says the footage can be 'selected' in post-processing, not just reviewed. There is no legal framework for what a camera that thinks rather than sees actually produces.
Sora burned $1M a day in compute costs. It made $2.14M total before OpenAI shut it down — and killed a $1B Disney deal in the process.
Australians use Claude 4.1x more than their population would predict — but coding tasks are 8 percentage points below the global average. Office and personal life use is what is driving the adoption. Their autonomy score: 3.38 out of 5. Collaborators, not delegators.
Disney announced a $1B investment in OpenAI and a Sora licensing deal. No money changed hands. Sora burned $1M/day. Disney found out it was dead 30 minutes before the public did.
The npm package included a source map anyone could read — and inside was a feature called Undercover Mode that strips AI attribution from code commits, a frustration detector watching user keystrokes, and a planned Tamagotchi pet for your terminal. The $2.5B product runs on the same kind of opera...
Anthropic is warning officials that its unreleased Claude Mythos model makes mass cyberattacks far more likely. The problem: nobody outside Anthropic can verify how it knows.
Every AI model you've trained ran on a number nobody ever tested. That might be about to change.
Turns out measuring AI chart skills is harder than measuring the charts themselves.