Uber burned through its entire 2026 AI budget by mid-April. Four months. Gone. In December, Uber rolled out Claude Code to roughly 5,000 engineers. Internal adoption took off so fast that the company set up a leaderboard ranking teams by total AI tool usage. Per-engineer API
@aihighlight
-
Mythos cybersecurity capabilities, not benchmarks, put it on White House watch list
By
–
Remember that Mythos isn't just a chatbot. The cybersecurity capabilities are what put it on the White House's watch list, not benchmark scores.
-
AI as a tool separates winners from losers
By
–
AI is a tool. The companies winning are the ones treating it like one. The ones losing are treating it like the strategy.
-
Agentic workflows automate outdated busywork
By
–
Half the agentic workflows shipping right now are automating busywork that should have been killed years ago.
-
Keep Using AI, Read Its Outputs While You Can Ask Questions
By
–
The lesson isn't to stop using AI. It's to read what it writes while you still can ask it questions.
-
Rhetoric shifted from automate to augment but nothing changed
By
–
Two weeks ago it was "automate most white collar work." Now it's "augment humans." Nothing actually changed except how they're talking about it.
-
Utopai Studios: One Prompt, Continuous Fantasy Film
By
–
The hard part is making that moment continue.
— AI Highlight (@AIHighlight) 29 mai 2026
So we tested @UtopaiStudios with a wizard-school fantasy film, the kind of world that usually needs a full cast, sets, costumes, voices, and serious production time.
One prompt. One continuous story.
Here’s what happened: pic.twitter.com/fbFQY5iAZzThe hard part is making that moment continue. So we tested @UtopaiStudios with a wizard-school fantasy film, the kind of world that usually needs a full cast, sets, costumes, voices, and serious production time. One prompt. One continuous story. Here’s what happened:
-
Decentralized Training Scales Video Models Without Data Centers
By
–
This is the part nobody is talking about yet.
— AI Highlight (@AIHighlight) 28 mai 2026
You can now train a serious video model without owning a single data center.
Bagel proved decentralized training scales from images to video, and the next stop is world models. https://t.co/vhPzDocy2BThis is the part nobody is talking about yet. You can now train a serious video model without owning a single data center. Bagel proved decentralized training scales from images to video, and the next stop is world models.
-
Meta Ray-Ban Glasses Footage Used to Train AI
By
–
BREAKING: Meta's Ray-Ban smart glasses record what you see. Some of that footage is watched by people. When a user shares data to improve Meta AI, the clips their glasses captured can be sent to human contractors who review and label them by hand. The reviewers work for a firm