Hey, AI Enthusiast!

Welcome back to the #1 AI newsletter on the planet!

Here’s what we have today:

Attention

Your Glazed Eyes Are Tipping Off AI

Unless you’re watching a long video that you’re really interested in, your mind has a high chance of wandering off.

When that happens, whatever’s being said goes in one ear and out the other.

But what you don’t notice is that your eyes start to make small changes.

These are things like how long they stay glued to one spot or how wide your pupils are.

A lot of times they’ll also glaze over, and that look gives you away.

Researchers are now teaching AI to read it.

They gathered years of studies on catching a wandering mind, then put AI models through every dataset they could get access to.

The goal was to see which ones spot it best.

And wouldn’t you know, out of everything they looked at, eye tracking turned out to be the most common way to do it.

It’s still research, and how well it works depends a lot on what setup is used.

Spotting the glazed look is one thing, but by the time anyone notices, you’ve already missed the part that played.

Normally, if you lose your train of thought while reading, you’ll go back and reread the lines you skipped.

But video doesn’t seem to work the same way.

A separate study followed college students watching their class videos, and after catching their own minds wandering, they almost never hit rewind.

A couple of them even typed out that they’d go back and rewatch it, then never did.

Think about the last long video you sat through.

There’s a good chance a chunk of it played while you were somewhere else, and whatever got said during that time, you never heard.

Our take: Everyone’s mind wanders at some point. What decides how much you lose when it does is the next few seconds, because going back means finding the spot and sitting through it again. Pressing on always feels easier. That’s how you get to the end of a video thinking you watched all of it, only to ask yourself what you just watched.

Together with Wispr Flow

Catch Up Without Hitting Rewind

Those few minutes you missed when you zoned out are still sitting back there in the video you just watched.

The only way to find what you missed is to scrub back through it until something sounds familiar.

That was until Wispr Notetaker came along.

Turn it on and it listens to whatever you have playing on your computer.

It picks up a long video, a podcast episode, an online class, meetings, or anything else with audio.

Here’s one way to use it.

Say you’re 40 minutes into a documentary and realize the last few minutes went right past you.

Instead of scrubbing back through the video, you can hit “What did I miss?” and it’ll catch you up on what was said.

If something didn’t quite land, type it into “Ask anything,” and you’ll get an answer pulled from what was said.

And if you want to take notes yourself while you watch or listen, there’s a “My thoughts” tab built in for that as well.

Robots

Humanoid Fights a Human in a Cage

Frankie LaPenna, a well-known content creator, has gone viral once again.

This time it wasn’t for taking work calls from a dirt bike behind a green screen.

Nope. This time it was for fighting a humanoid robot.

Normally, REK puts on live events between humanoids, but this time it put LaPenna in the cage against three of them.

They billed it as the first ever human-versus-humanoid cage fight.

You can easily find the clip online.

What you’ll see is LaPenna throwing punches.

But the robot isn’t having it and throws its own punches and kicks, with one of them sending him flying.

What the clip doesn’t show is who’s behind the robot.

A human was controlling the robot the entire time, with VR gear or a gamepad.

AI handled the balance and the trained moves, but every punch it threw came from someone holding the controls.

LaPenna wanted to go in without protective gear, but REK made him wear motorcycle-style gear and gloves.

And even with gear on, he still hurt his hand throwing a punch.

The fight didn’t end with LaPenna getting knocked down, though.

He got back up, knocked the robot off balance, and finished the fight with the robot’s head coming off.

Our take: Look past the flying kick. The part worth noticing is that the robot mostly stayed on its feet. A machine that takes a hit, catches its balance, and swings back, with AI keeping it upright the whole time, is a big jump from the wobbly walking robots we saw only a few years ago. In time, these humanoids are only going to get better and better.

Pets

Your Cat’s Face Shows When It Hurts

Cats are great at hiding pain.

And if something ever happens to them, a lot of owners have no idea until it’s already gotten worse.

But if you know where to look, their faces still give it away.

Vet researchers took those signs and created a simple checklist called the Feline Grimace Scale.

Then, they trained AI to look at a photo of a cat’s face and give it a pain score using that same checklist.

You don’t need the AI to spot these changes.

Here are the telltale signs.

Each one gets a 0 = looks normal, 1 = you’re not sure, or 2 = the change is obvious.

Watch your cat for 30 seconds while it’s calm and awake, add up the scores, and anything at 4 or more out of 10 is worth a call to the vet.

It may be tempting to send a chatbot a photo of your cat and ask if something’s wrong.

Well, researchers did that and what they found out is that the chatbots (older versions, to be fair) rated cats as less painful than a vet did.

The catch: The scale is built for sudden pain, like after an injury or surgery. Long-term pain like arthritis shows up more in how your cat moves, jumps, and eats.

Our take: A lot of pet health questions start with “she seems off.” That doesn’t give a vet much to work with. Having a score is something concrete you you bring to the vet, and checking the same face again after the visit tells you whether things are getting better.

Coming in Hot

AI Tools of the Day

Farming

The Letter Behind a Melon’s Flowers

Every plant runs on a DNA code millions of letters long, and one swapped letter is sometimes all it takes to change how a crop grows.

Figuring out which letter that is has been eating up years of plant breeders’ time.

When they compare plants with and without a trait, they end up with a huge list of possible mutations.

Narrowing it down usually means growing thousands more plants and running greenhouse tests for many seasons at a time.

The reason it takes so long is because DNA gets passed down in big chunks.

A harmless mutation sitting next to the important one gets inherited right along with it. On paper, the two look identical.

An AI biology lab tried a different way in.

It paired Google’s open model, Gemma, with BOTANIC-1, a model trained on the DNA of hundreds of plant species.

Gemma handles the conversation and runs the analysis, while BOTANIC-1 reads the DNA itself.

Its trick is evolution.

If a spot in the DNA has stayed the same across millions of years of plant history, a change there is likely a big deal.

So the lab picked a melon mutation researchers had already found to test it.

That was one letter that switches the plant’s flowers from female to having both male and female parts.

Gemma narrowed the list down to two mutations that scored exactly the same on its own, the real one and one in an unrelated gene.

Once BOTANIC-1 was added, the real one came out on top out of 2,494 candidates.

The catch: The mutation was already known, so the test proves the method works rather than finding something new. The research is also a preprint that hasn’t been peer reviewed yet.

Our take: Breeding a crop that handles drought or heat has always been slow, because the answer sits somewhere in a list nobody can rule out quickly. If this holds up on mutations nobody has found yet, years of ruling things out turn into a short list worth testing first.

Interesting AI

This Robot Cooks Your Dinner

After you’ve worked all day, the last thing you want to do is get home and have to cook dinner.

Standing over a hot stove, stirring, and waiting to add the next ingredient isn’t considered fun for most.

Posha wants to make dinner easy by doing it for you.

It’s a countertop robot that comes loaded with more than 1,000 recipes.

You pick one, give it fresh ingredients and spices, and walk away.

The built-in camera keeps an eye on things the way a chef in the real world would.

And the app lets you watch from another room, swap out ingredients, or increase the amount of spices used if you want.

After it’s done cooking, you set the table, eat, and all that’s left is to put the parts in your dishwasher.

Smart Glasses

AI Glasses Might Get Banned

AI glasses have been getting a bad rep lately.

That’s because the person next to you might be wearing a pair that’s recording everything you say and do without you even knowing.

That worry is pushing Norway to propose a temporary ban on AI glasses in public places.

It’s the first major country to try it, and the list of places it’s looking at covers a lot of everyday life.

Parks, beaches, museums, shopping centers, gyms, and public events are all on the table, along with schools, kindergartens, and playgrounds.

But it’s not a total ban.

People would still be able to use the glasses privately and in other places.

The government also hasn’t decided which devices count yet.

Glasses with a camera and a microphone are on the list, and so is anything else you wear that records or runs AI.

Companies are drawing their own lines as well.

One major cruise line has kept smart glasses out of its ships’ public areas since July.

Our take: It used to be easy to spot a camera, especially when it was pointed at you. Even a phone pointed at you could be recording. The difference is you know what’s happening. Someone wearing what looks like an ordinary pair of glasses is a different story, and that’s why the rules are being written for the people around the wearer.

Prompt of the Day

Fool Me Twice

Important Note: You’ll need to click the button to get the complete prompt.

That’s all for Monday!

Thanks for reading,

— There’s An AI For That