AI Made a Discovery, Rebuilt Your Feed and Reached for Space

Instead of writing code or answering questions, around 950 Claude agents were asked to search through a huge biological dataset for unusual reverse transcriptases.

They worked for roughly 21 hours, processed around 210 million tokens, and examined more than 200,000 possible candidates.

Eventually, the system spotted an unusual repeating DNA pattern connected to a previously uncharacterized biological system now called ART.

Scientists then took that result back into the laboratory for further investigation.

The important part is not that Claude suddenly became a scientist on its own. Human researchers still designed the task, checked the results, and are still trying to understand what the system actually does.

What is new is the role AI played.

Claude did not simply summarize existing research. It searched through a massive biological space, found an unusual pattern, and gave scientists something worth testing.

That may be an early example of how AI changes scientific discovery: not by replacing researchers, but by searching through possibilities far faster than humans realistically could.

Read More → Anthropic

Now YouTube wants to give users another option.

Its new Custom Feeds feature will let people describe the kind of content they want to see and generate a separate feed around that request.

You could ask for something specific like:

“30 minute documentaries for my evening commute.”

Instead of waiting for the algorithm to learn that preference over time, YouTube can create a dedicated feed around it.

Users will also be able to save multiple Custom Feeds for different interests or routines.

It sounds like a small change, but it flips the usual recommendation model around.

Instead of the algorithm constantly guessing what you want, you can simply tell it.

For now, YouTube says the feature will begin rolling out in the United States, but the larger idea could have much wider implications for how recommendation systems work across the internet.

Read More → YouTube

Google is preparing to test something that sounds more like science fiction than cloud computing: putting AI hardware into orbit.

Under Project Suncatcher, Google plans to send Tensor Processing Units into low Earth orbit to see whether the same chips used for AI workloads on Earth can survive launch vibration, radiation, extreme temperatures, and cooling in space.

The experiment itself is small, but the idea behind it is much bigger.

AI data centers consume enormous amounts of electricity, and Google is exploring whether future computing clusters could take advantage of near continuous solar energy in orbit. The company is not launching a full data center yet. This first mission is about finding out whether the hardware can work reliably outside Earth.

If the test succeeds, the next step could involve linking satellites together so they can share AI workloads.

That raises a fascinating question:

Could some of the infrastructure powering future AI eventually move off the planet?

Read More → Reuters