Vegas clears ~7,000 robotaxis, Nvidia eyes a chip startup, and YouTube writes you a new song

Nevada has cleared Las Vegas for the biggest driverless rollout in the US so far: up to 5,000 Tesla robotaxis, plus 1,000 each for Waymo and Uber, running with no safety driver and real paying passengers. The technology is genuinely impressive. The real test is the trust jump to thousands of unsupervised cars sharing public streets.

The problem

Robotaxis have spent years stuck in tightly limited pilots, small fleets in a few blocks, always with a human safety driver ready to grab the wheel. Scaling to a full city with nobody behind the wheel has been the barrier nobody wanted to be first to cross.

What changed

On Aug 21, Nevada regulators approved permits allowing Tesla to deploy up to 5,000 robotaxis across the Las Vegas area over the next year, with Waymo and Uber each cleared for fleets of up to 1,000 vehicles. All of it is driverless, paid, and on public roads, making Vegas the largest driverless testbed in the US.

The catch

Approval is a ceiling, not a headcount. These fleets scale up over the coming year, not overnight, and 'up to 5,000' means the real number on day one is far smaller. Thousands of unsupervised cars also means the safety record now gets written in public, at scale, for the first time.

What it means for you

If you visit Vegas in the next year, expect to be able to hail a car with nobody in the front seat. Watch how the first few months go: this rollout is effectively the industry's real-world stress test for whether driverless-at-scale is ready.

Bloomberg reports Nvidia is in early talks to invest in, or acquire, Rebellions, a South Korean startup that builds power-efficient AI inference chips on Samsung's foundry. Nvidia already owns AI training; inference, the job of actually running models, is the next battleground, and that is exactly Rebellions' specialty. It is still a leak, not an official deal.

The problem

Nvidia dominates the chips that train AI models, but training is a one-time cost. The bigger, recurring bill is inference, running those models for billions of daily queries, and that market is wide open to rivals building cheaper, more power-efficient silicon.

What changed

On Aug 21, Bloomberg reported Nvidia is in early-stage talks with Rebellions, a 2020-founded South Korean firm specializing in AI inference NPUs built on Samsung's foundry with HBM3E memory. The talks reportedly range from technology licensing to equity investment to a full acquisition.

The catch

This is a report of early talks, not a signed deal, so treat it as a leak rather than fact. Any acquisition of a foreign chip designer would also face regulatory scrutiny, and Nvidia buying an inference specialist could invite antitrust questions given its dominance.

What it means for you

Watch this as a signal, not a product. It tells you where the smart money sees AI heading next: away from just training and toward cheap, efficient inference, which is what eventually decides how much AI features in your apps and phones actually cost.

On a Mac, to grab a file's full path, right-click it in Finder and then press and hold the Option key. 'Copy' quietly changes to 'Copy [file] as Pathname.' Paste that exact path straight into Terminal, an upload box, or a support ticket, no dragging the file around and no typing the path by hand.

The problem

You constantly need a file's exact location, for a Terminal command, an upload dialog, or a bug report, but macOS hides full paths by default. Most people resort to dragging the file into a window or typing the path out and getting it wrong.

What changed

Finder has a hidden modifier: right-click any file, then hold Option. The 'Copy' item transforms into 'Copy [file] as Pathname,' which puts the complete absolute path on your clipboard instantly. It has worked this way for years and almost nobody knows it exists.

The catch

It only appears while Option is held, so if you right-click and don't press the key you'll never see it. It copies the path as text, not the file itself, so don't use it when you actually mean to copy the file to paste elsewhere.

What it means for you

Next time you need to point Terminal or an upload box at a file, right-click, hold Option, Copy as Pathname, and paste. It's a five-second habit that removes a daily annoyance.

If a video gets a Content ID claim, YouTube Studio on desktop now has a 'Create' button in the Replace Song tool that generates four royalty-free instrumental tracks you can swap in without taking the video down. It's rolling out to US desktop users first, with wider and mobile support later. Honest take: it's a great way to save a flagged upload, not a replacement for real music.

The problem

A single copyrighted track can trigger a Content ID claim that redirects a video's revenue or blocks it in some countries. Creators' usual fixes, muting the audio or pulling the video down, either ruin the content or erase its momentum.

What changed

YouTube added a 'Create' button inside the Replace Song tool in Studio (desktop). It generates four royalty-free instrumental options you can drop straight into the flagged section, releasing the claim while the video stays live and keeps its views and comments.

The catch

It's limited to desktop users in the US for now, with a broader and mobile rollout promised later in 2026. The tracks are generic AI instrumentals meant to resolve claims, not to match your video's mood, so it's a patch rather than a soundtrack upgrade.

What it means for you

If you post to YouTube, keep this in your back pocket for the next claim: open the video in Studio on desktop, Replace Song, Create, pick a track, and your upload stays online instead of going dark.

Every phone, app, and gadget now advertises itself as 'AI-powered.' Honest question, no wrong answers: in your daily life, is AI genuinely making your tech better, or is it mostly a marketing sticker? Reply with one AI feature you actually use and would miss, and one that's pure fluff.