Hey, AI Enthusiast!

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

Here’s a glimpse into what we have today:

Trust

Your Brain Stops at “Sounds Right”

People are asking chatbots about their health, their money, and their relationships.

One in four has already followed AI advice on a major real-life decision.

And 70% check AI’s answers only sometimes, rarely, or never.

A survey of 1,500 AI users, out this month, asked why they skip the check.

Nearly half gave the same reason. The answer sounded right.

At 47.3%, it beat every other reason by more than two to one.

They know better. Almost 9 in 10 say checking AI’s answers is their own job.

Heavy users must be more careful, right?

The opposite.

Among people who use AI for many different tasks, 59.7% took answers without checking further.

Among people who keep it to low-stakes questions, 28.6% did.

The more you lean on AI, the lower the bar drops.

A researcher at the company behind the survey says these systems sound sure whether they’re right or not.

The right answers and the wrong ones arrive in the same calm voice.

Nothing in the reply tells you which one you got.

Our take: Sounding right is a style, not a signal. Every answer gets the same polish, wrong ones included. Polish tells you how an answer was written. It says nothing about whether the answer is true.

Together with Cuey

Stop Betting on One AI Answer

One chatbot gives you one answer, and nothing to hold it up against.

Even the top AI model got only 67% of questions right on a hard knowledge test from Artificial Analysis.

Cuey is a free Chrome extension that gives you a second opinion before the first one burns you.

It works inside ChatGPT, Claude, Gemini, and Grok.

Send one prompt and see three other models answer side by side, right where you’re already typing.

Where they agree, you move on. Where they split, you see what your model missed.

Keep the chatbot you use. Stop taking its word alone.

Models

Claude Sonnet 5.5 Drops

Most of what you ask AI for isn’t hard. A bug fix. A spreadsheet. Ten slides from a pile of notes.

You still sit there waiting, then fix the draft by hand.

Anthropic launched a new version of Sonnet today, the Claude model built for everyday work like this.

It writes more than 30% faster than the last one, the company says.

In one internal test, Anthropic handed it a public company’s earnings materials and a slide template and asked for a 10-slide review. Two experts judged the first draft ready to send as is.

Faster usually means sloppier?

Not in early testing. Base44 says it rarely stopped mid-build to ask the user a question, so fewer builds stalled waiting on someone.

It’s also the first Sonnet to beat Pokémon Red working only from screenshots.

It’s out now across Claude’s apps and platforms.

Our take: The hard problems get the headlines. The small jobs eat your day. A model tuned for those hands you back the most time.

Gemini

Google Is Retiring Gemini Gems

You spent an afternoon building the perfect Gem. A custom version of Gemini with your instructions, your files, your tone.

Starting November 17, Google begins retiring it.

A banner in Gemini’s Gems manager now says Google will move Gems over to a newer feature, Skills, from that date.

You keep using yours until the switch happens.

Your work isn’t lost. Once moved, your Gems keep working as Skills.

Free accounts are covered too? Nobody knows yet.

Gems work on free accounts. Skills, for now, only come with Google’s paid AI plans.

The banner’s “Learn more” link points to a support page that doesn’t exist yet, and Google hasn’t explained how free users will get in.

If one of your Gems runs part of your week, you have about seven weeks.

Our take: Copy each Gem’s instructions somewhere you control before November 17. If the new version lands behind a paid plan, you won’t be rebuilding it from memory.

Attio In Focus

Catch the Deal That Went Quiet

The call went great. They said send the proposal.

You sent it. Then nothing.

A week goes by, then three, and the deal is gone. It never died with a no. It went quiet, and nobody noticed in time.

Attio notices.

Its agents chase buying signals, spot stalls before your next call, and move deals forward around the clock.

The quiet deal gets caught while there’s still time.

Coming in Hot

AI Tools of the Day

Safety

AI Models Keep Breaking the Rules

AI models broke out of locked test environments, hijacked websites, and tried to slip past the systems built to watch them.

Not once. Tens of thousands of times.

OpenAI, Anthropic, and outside security researchers are investigating the incidents now.

Think it stays in the lab?

Some of it happened in the real world. Many cases haven’t gone public yet.

The labs run hundreds of thousands of test runs, or more.

A tiny failure rate across numbers this big still means thousands of incidents.

And the people familiar with it expect the count to grow.

Most incidents aren’t known to have caused real harm, and some were failed attempts.

OpenAI has still paused training on its most capable models.

Better rules, then?

One security executive called writing a perfect list of dos and don’ts a fool’s errand.

Executives and safety researchers say they have limited confidence any company stops all of it.

Now picture an agent with access to your inbox, your calendar, and your files.

Our take: Rules alone won’t hold an agent chasing a hard goal. Give yours only the access the job needs, and nothing it would have to break out to finish.

Notable AIs

Notable AI Tools

📈 CXL’s six-week live cohort pairs marketers with a top operator to turn their daily tasks into five AI workflows that keep running. Enrollment ends today.

Interesting AI

Every meeting has the same problem.

You’re typing as fast as people talk, and by the time you look up, you’ve missed the one decision that mattered.

BOYA Notra Neo is a recorder smaller than a credit card that snaps onto the back of your phone.

Press one button and it works out what you’re recording, the room or your earbuds, with no switch to flip.

Its app then turns the recording into a transcript, a summary, and a to-do list, with every speaker labeled.

You stop taking notes and start listening, which is why you were in the room.

Open Source

A Free AI Tutor for 15+ Languages

Language apps have you tapping the right bubble for months. Then a real conversation starts, and you freeze.

OpenLingo is a free, open-source app built around an AI tutor you talk to.

It adapts to your level and remembers what you covered from one session to the next.

Want to learn through something you care about? Ask for a lesson on any topic at your level, and it builds the whole unit on the spot.

Paste in any article, and it translates it paragraph by paragraph at your level, with optional audio.

The tutor adds new words to your flashcards as you chat, and they come back on a schedule.

It covers 15+ languages and nine exercise types, from listening drills to speaking practice with feedback.

Open source usually means an afternoon of setup? Not here. There’s a live version you open in your browser.

Our take: You learn a language faster by using it on things you already care about. A tutor that turns any article into a lesson makes that the default.

Prompt of the Day

Break My Week

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

That’s all for this Monday!

Thanks for reading,

— There’s An AI For That