Hey, AI Enthusiast!

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

Here’s what we have today:

Support Bots

Most People Have Yelled at a Bot

You call for help. A bot answers.

You explain the problem. It answers a different one.

You try again, slower this time. Same answer.

So you yell “Human” into your phone.

In a recent survey, six in ten Americans said they’d yelled at a bot to get past it faster.

People haven’t given up on calling, either.

When something goes wrong, picking up the phone is still the first move for more people than any other option.

That’s the problem. The channel people trust most is the one failing them most.

Only 7% say automated phone lines consistently fix their issue.

The survey also asked what frustrates people most about support.

The top answer: a bot that doesn’t understand them. It beat long hold times and getting bounced between departments.

And people don’t give a bot many chances.

Most will repeat themselves once. If it still doesn’t get it, they’re gone.

The frustration follows them off the call, as well.

More than half admitted to crying, yelling at someone they love, or breaking a device after a bad support experience.

You can sit through hold music. You can’t get anywhere with a bot that keeps getting you wrong.

Our take: Nobody’s yelling at the bot for being a bot. They’re yelling because it didn’t understand them. Most people say they’d keep using a bot if it got things right. Most bots haven’t earned that yet.

Together with Speechmatics

One Misheard Name. One Lost Caller.

A lot of the time, the bot didn’t misunderstand you. It misheard you.

You said “Okafor.” It heard “Oak Ford.”

Everything after runs on the wrong word.

Some callers won’t even give it the chance. Nearly a third say they’d hang up the moment AI answers.

The ones who stay won’t give it many tries. And the speech layer slips on exactly what they say (names, card numbers, accents, a noisy kitchen in the background).

Linden by Speechmatics fixes the listening, so the rest of the call finally has the right words to work with.

It’s speech-to-text built for voice agents, made for the accents, names, and numbers general models trip on, in 55+ languages.

Say anything in your own accent and watch what it catches.

Health

20s of Speech Flags Diabetes Risk

Type 2 diabetes can go years without a single symptom.

Catching it takes a blood test. Plenty of people skip it until something’s already wrong.

Now researchers think your voice might give it away first.

London startup thymia and RMIT University in Melbourne trained an AI on more than 63,000 voice recordings from about 21,000 people.

Everyone read the same short fable out loud. Twenty seconds each.

Same words every time.

The Al wasn’t listening to what people said. It was listening to how they sounded.

And it heard plenty. Checked against blood tests, it caught 82% of the people who had diabetes.

The results were presented at a European diabetes meeting, but nobody has checked them independently yet.

Our take: A 20-second voice check could reach people who’d never book a blood test. For now, a flag might be a reason to get one, but don’t treat it as a diagnosis.

TAAFT Update

TAAFT Video Generator

Our TAAFT Video Generator now supports image to video.

Upload a picture, describe what should happen, and get a video back.

You can also create videos longer than 8 seconds now.

Pick “Fast” for the quickest result or “Quality” for richer motion.

And you can generate up to 100 videos a day.

Generated content is 100% free to use, including commercial use.

Travel

AI Is Coming to Air Traffic Planning

Sitting at the gate. Sunny outside. Flight’s delayed anyway.

The holdup is usually somewhere else in the sky, and someone has to untangle it.

That person is an air traffic manager.

They plan how flights move across the whole country, and the CEO building their new tool says the map app on your phone predicts traffic better than the software they’ve been using.

So the FAA is going to hand them AI. It comes from Air Space Intelligence, a Boston company.

It pulls from more than 200 sources, including weather, congestion, turbulence, winds, and runway closures, and works out the best path for every flight in the country.

The CEO calls it chess.

The software helps managers see four or five moves ahead, so traffic keeps moving and fewer flights get delayed or canceled.

The AI doesn’t fly anything, and it doesn’t keep planes apart. Controllers still handle it.

It makes a suggestion. A person decides.

A version for airline dispatchers already runs at Alaska Airlines and other carriers.

The FAA’s starts at three big airports around Washington, D.C.

Our take: The people planning the country’s flights deserve tools at least as smart as the map app in your pocket. Judge this one by whether your next delay gets shorter.

CrafterQ In Focus

Your Website, Answering Back

A visitor lands on your site with one question. They click around, can’t find it, and leave.

The answer was there the whole time, buried three pages deep.

CrafterQ turns your site, docs, FAQs, and files into an AI agent that answers every question, then points visitors to the next step (a demo, a quote, a purchase).

It goes live on any site with one line of code.

Coming in Hot

AI Tools of the Day

AI Detection

Google Explains Its Hidden AI Watermark

That volcano clip in your feed might be fake. And the tools built to tell you whether or not it is are losing.

AI detectors work by spotting tells. Every new model leaves fewer of them.

On the Google DeepMind podcast, they talk about SynthID, which skips the hunt and plants a mark instead.

Images, video, and audio from Google’s models carry an invisible signal. Crop it, rotate it, zoom in, add noise.

The mark is built to survive all of it. Text is trickier.

A sentence can usually go a few ways, so a secret key quietly nudges which words get picked.

One word proves nothing. A whole paragraph of nudges gives it away.

Short answers slip through. “The capital of France is Paris” leaves no room to hide anything.

The catch: it only spots content from Google and partners who share their keys. Anything else gets no verdict at all.

Our take: Spotting fakes by eye is a losing game for now. Make checking a habit before you share, and treat “no mark found” as unknown, not real.

Notable AIs

Notable AI Tools

💬 Attio talks through your deals and accounts in chat, then updates the stalled one and sends the follow-up after your go-ahead.

🔌 Nango lets the AI agents in your product act on and pull data from 1,000+ APIs, with auth, tool calls, syncs, and triggers handled for you.

History

An Amateur Cracked a Napoleonic Cipher

For 217 years, nobody could read this letter.

It’s a coded war briefing to one of Napoleon’s generals, and it sat on the list of unsolved historical ciphers.

The only copy was one blurry scan in a 1969 French army journal. With the wrong date on it.

A few weeks ago, an amateur used AI to read a WWII Enigma message.

Now Carter Church, who had never broken a cipher in his life, has done the same with Napoleon.

OpenAI’s GPT-6 Astra did the heavy lifting in about six hours.

It cut the image into rows, read 1,300 hand-drawn symbols, worked out which ones were the same mark written twice, and broke the code.

What came out was where every French army stood, written about two weeks before Austria invaded Bavaria in 1809.

Napoleon’s own printed order for this letter breaks off mid sentence, right where he says what his general shouldn’t be held back by.

A row of dots. Nobody knew what they were hiding.

The decoded letter carries the same instruction with no gap, “a few troops or a gathering of rabble.”

To rule out the AI reciting history it’d already read, he reran it with every Napoleonic text stripped out. Same answer.

Our take: These letters weren’t unsolved because they were impossible. It’s only because of the amount of time it would take to solve one old page. Now an evening does it.

Prompt of the Day

Talk Me Out Of It

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

That’s all for Thursday!

Thanks for reading,

— There’s An AI For That