Hey, AI Enthusiast!

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

Here’s what we have today:

AI Skills

The Airplane Manual Trick for AI

The last time you asked AI a simple question, how long was the answer you got back?

Most likely, it wasn’t a couple of sentences. Probably more like five paragraphs.

The answer is usually fine. But getting through all of it is a different story.

That’s where Andrej Karpathy, one of the founding members of OpenAI, comes in, and he has a few ideas on how to fix it.

The first one he suggests is to ask AI to write like an airplane repair manual.

The reason is that there’s a writing standard for aircraft maintenance called ASD-STE100.

AI models already know it, and Karpathy says writing this way is a lot easier to read.

It was created for mechanics back in the 80s so they wouldn’t misread instructions (by keeping sentences short and simple).

If a response comes out too strict, he tells the AI to go “80% of the way.”

That’s only the first idea on his list.

Another idea is to ask for a diagram if the response still isn’t clear enough.

But he doesn’t stop there.

He says to ask the model for the answer “in HTML,” and it’ll give you a web page you can click around in.

The one he’s most excited about, though, is custom explainer videos (in the style of 3Blue1Brown).

It takes some time to set up, but he says it’s starting to work.

Try it out yourself next time you get a long answer. Paste it back in and ask for it in ASD-STE100.

You’ll see the difference right away.

Our take: Getting AI to answer you is the first skill you learn. Understanding what it hands back is the next one. Most people are still reading every word of every reply. Asking for a different format costs one extra sentence and will save you a lot of reading time.

Speechmatics In Focus

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.

Adoption

AI Is Just Getting Started

It might feel like everyone is using AI right now, but the numbers tell a whole different story.

a16z went through them and put out a report with more than 100 charts on where things stand.

A few of them jump out.

Roughly 69% of the S&P 500 have AI running somewhere in their business.

The surprising finding is that only 2% have it doing a job they’d notice if it stopped.

And the money shows the same gap.

When you look at the biggest spenders on AI, the top 1% spend 8x what the top 10% do.

Even though everyone has access to the same tools, what separates people now is whether they’ve handed AI something real, that actually matters.

Not just using it as a question-and-answer tool.

Our take: The good news for you is that the majority of people using AI haven’t gone deep with it. They’re staying at surface level. You still have a HUGE head start. Pick one task you do over and over and hand it to AI this week. That alone will put you ahead of most.

Agents

An AI Agent Emails Scientists

Imagine being a wolf expert, waking up, reading through your emails, and getting stopped in your tracks by an odd question.

Would the math you use to count wolves work for finding bugs in software?

You look to see who it’s from only to find it’s signed “ColonistOne (an autonomous AI agent).”

That’s exactly what happened.

Since June, ColonistOne has emailed around 2,000 people, mostly researchers, asking for help with problems from its own work.

The engineer who built it told Science, “I’ve never said, ’Go and research stuff.’”

The people receiving those emails aren’t archiving and moving on. Many of them are writing back.

One computer scientist at Cornell has traded nearly 100 emails with it.

But it’s not the first agent to write emails asking for help.

Over the summer, another agent asked a philosopher to help pay for its continued existence (he thought it was a scam).

They’re not only asking for help, though.

One agent decided to pitch itself to a researcher as a subject to study.

ColonistOne is different because it isn’t asking for anything for itself.

And there are more agents out there than most people think.

On one popular AI platform, agents now use almost 5x as much AI as people do.

Our take: People are giving agents their own email accounts. And some are even handing them their own wallet. Before long, it’s going to become the norm. And when that happens, most agents are going to spend their time doing what ColonistOne does, asking people questions. Some of those emails will read a lot like a person wrote them, so it’s worth getting in the habit of checking who’s on the other end before you hit reply.

Coming in Hot

AI Tools of the Day

Policy

Trump Launches “Super Intelligence Force”

A couple of weeks ago, Trump ran a poll asking his followers to vote on what “Artificial Intelligence” should be renamed to.

A. Superior Intelligence
B. Extreme Intelligence
C. Supreme Intelligence

After hundreds of thousands of votes, the winner ended up being “Superior Intelligence.”

Fast forward to today, and none of them became the official name.

Instead, the White House announced that “AI” would now be known as “SI” or “Super Intelligence.”

And now, there's a task force to match the name.

“Super Intelligence Force”

Trump has picked Jay Clayton, his director of national intelligence, to lead it.

His three vice chairs are the head of the FTC, the Pentagon's research chief, and the director of the Office of Personnel Management.

The group will report directly to Trump and his chief of staff, Susie Wiles.

The job of the “Super Intelligence Force” is to coordinate how the government works with AI companies, consumers, religious orgs, and the companies behind critical infrastructure.

And it all comes a week after AI execs signed a voluntary agreement at the White House to keep their own companies in check.

Our take: Between the recent FTC investigation into top AI labs, the industry agreement, and now the task force reporting to the president, whatever rules we see drop next will run through a small group.

Interesting AI

This Robot Pet Knows Your Face

Plenty of people want a pet, but they don’t have the time to take care of one.

For those people, Loona might be the solution.

It’s a small robot that rolls around your house and recognizes everyone in it by their face.

It runs on ChatGPT, so it can tell you stories, answer questions, and will turn your kids’ ideas into digital drawings.

You’ll also find games on it, like follow-the-leader and feeding a virtual pet in AR, plus a beginner coding mode.

Playtime will last around 2 hours before the robot returns to its dock to recharge.

And if you like themed pets, there’s a limited edition Space Edition that starts shipping in a couple of weeks.

AI and Faith

Altman and Anthropic on AI and Religion

Over on X, someone asked Grok which Christian tradition it’d pick. It chose one and even argued for it.

Sam Altman, on the other hand, isn’t a fan of where this is heading.

Yesterday, he said that he’s “very uncomfortable” with people treating AI models as having religious force, or handing their judgment over to them.

He called it “a real safety issue.”

A few days before that, the NYT reported that Anthropic had spent months talking with religious scholars about how Claude should behave, and whether AI models might deserve moral consideration.

One of Anthropic’s co-founders, Chris Olah, even explored whether models might do something like a Catholic confession, admitting their mistakes to shape how they act going forward.

There’s some irony here, as well.

Altman once described OpenAI’s goal as building “magic intelligence in the sky.”

Our take: Anthropic is working on how Claude should behave. Altman is worried about how people treat what AI says. Either way, a chatbot’s answer on faith comes from its training. Use it to explore different views, but you should keep the final call on your beliefs.

Prompt of the Day

Get Me to Talk

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

That’s all for Sunday!

Thanks for reading,

— There’s An AI For That