Hey, AI Enthusiast!

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

Here’s what we have today:

Memory

The Notes ChatGPT Keeps on You

Go to ChatGPT, start a new chat, and ask what it remembers about you.

There’s a good chance it’ll bring up things you forgot you ever told it.

A good example is a trip you were planning for New York earlier in the year and the top things you decided to do during your trip.

And that’s a great feature.

But what happens when you have a long chat with it?

Somewhere along the way, it loses track of something you’d decided back in the beginning of the conversation.

That’s AI’s selective amnesia, and it comes down to how AI models read.

The way it works is that every word in a chat gets weighed against every other word in it.

Ten words means roughly a hundred of those connections, and a thousand words means about a million.

The longer a chat runs, the more it starts to lose attention.

It’s a bit like trying to keep too many things in your head at once.

People who build AI even have a name for it (context rot).

It doesn’t fade evenly, either.

Researchers found models hold on best to whatever’s at the start or the end of what they’re given.

And they hold on worst to what’s buried in the middle.

That’s why you should put what matters the most at the top or the bottom of your messages.

Back to what it does remember about you, that comes from a notebook.

As you chat with ChatGPT, it saves details about you and pulls them back in when they’re useful.

That’s how a brand-new chat still knows your job, your projects, and how you like your answers written.

As soon as you switch apps, though, the amnesia sets in.

The notes ChatGPT took on you stay inside the app.

Open a different chatbot like Gemini, and you become a total stranger.

There’s a workaround, sort of.

Claude and Gemini both have a way to copy your notes over.

Don’t expect those notes to sync between your chatbots.

Google even says there’s no ongoing sync.

Anything you tell one app after you copy them over never makes it back to the other.

Our take: When a chatbot seems to know you, it’s working from notes it rereads. Notes are the easiest thing in the world to copy, which makes the wall between apps stand out even more. And the better each app gets at remembering you, the more you leave behind every time you try a different one. That leaves you stuck with one app or copying your notes over again and again.

Together with Cuey

Switch Chatbots, Keep Your Context

Moving to a different chatbot shouldn’t mean that you have to introduce yourself all over again.

But that’s how it goes.

Open Claude after months in ChatGPT, and you’re back to explaining your role, your project, and how you like your answers.

Cuey takes care of that part.

Cuey carries your context across ChatGPT, Claude, and Gemini, so every model already knows who you are and what you’re working on.

Say you’ve spent weeks planning a kitchen remodel in ChatGPT and want Claude’s take on one section.

Open Claude, and it already knows the budget, the room, and the look you’re going after.

Your data stays yours, as well.

Cuey encrypts sensitive info on your device, never stores it on its servers, and doesn’t train on your prompts.

Neurotech

AI Pulls Pictures Out of People’s Brains

Researchers put people in a brain scanner, showed them photos, and then had AI redraw each picture using nothing but their brain activity.

You read that right.

The redrawing comes from the same kind of AI behind image generators, trained on brain scans paired with the pictures people were looking at.

Another team showed people short video clips, then rebuilt the videos from only their brain activity.

A team at UT Austin went after words instead.

They played people a podcast and rebuilt what they heard, sentence by sentence, keeping the meaning rather than the exact wording.

It even works on thoughts nobody says out loud (basically mind reading).

Tell yourself a story in your head, and the AI writes out a paraphrase of it.

But there’s a big catch.

Every one of these happened inside a lab scanner, which isn’t exactly something you’ll have in your living room.

On the other hand, you could think about what you want AI to do in the future and it just happens without you having to say or type anything.

Our take: This runs on the same trick as the chatbots you use every day, which is feeding a model enough examples until the patterns show up. Brain scans used to be too noisy for that, and AI is what changed it. Whether a device at home, that reads through your skull, picks up a clean enough signal to pull it off is the part nobody knows yet.

Images

Nano Banana Lets You Draw the Edit

If you’ve ever asked AI to change one small thing in an image, you know how it goes.

Sometimes it fixes what you asked for and you move on.

Other times it redraws half the picture along with it.

Well, Google’s newest version of Nano Banana goes after that problem.

Here’s what Google says got better.

You can find it in the Gemini app and in AI Mode on Google Search (tap the banana icon under the search box in the Google app).

If you prefer to use Google AI Studio, it’s there as well.

The catch: Google’s own notes admit small text still comes out blurry a lot of the time, and a character won’t always match perfectly from one image to the next.

Our take: Plenty of people give up on an AI image after they’ve repeated “keep everything, change only this” multiple times. Pointing at the exact spot you want changed is how you’d give notes to a designer, and it’s the thing that ultimately gets you what you’re after faster.

Coming in Hot

AI Tools of the Day

Movies

Sam Altman’s Firing Became a Movie

You probably remember when Sam Altman was fired from OpenAI back in November 2023.

Within a few days, he was already back running the company.

It was one of the craziest weeks tech has had, and now it’s a movie.

It’s called “Artificial,” and it premiered at the New York Film Festival with Andrew Garfield playing Altman.

The rest of the cast reads like an OpenAI org chart.

Yura Borisov plays co-founder Ilya Sutskever, Monica Barbaro plays Mira Murati, Cooper Hoffman plays Greg Brockman, and Ike Barinholtz plays Elon Musk.

Garfield says he spent between two to three months watching footage of Altman to get him right.

His first impression, from watching Altman testify before Congress, was that he seemed honest and transparent.

Then, he said at the premiere, you learn more about a person.

Getting it into theaters wasn’t easy, though.

Amazon MGM picked it up first, then dropped it in June, reportedly over its ties to OpenAI.

Netflix and Focus Features both passed before Neon stepped in.

So far, the early reactions out of the festival are strong, with one critic calling it a pitch-perfect dark comedy about the absence of responsibility.

It hits theaters on Christmas Day.

Our take: Most people never read a word about OpenAI’s boardroom fight, but plenty of them will see this movie. Whatever version of Altman ends up on screen is the one a lot of people will remember, the same way The Social Network shaped how people think of Mark Zuckerberg. Treat it as one version of events on what happened.

Interesting AI

Your Own Personal AI at Home

The more an AI assistant knows about your life, the more useful it gets.

The problem is that everything it knows usually ends up on some company’s servers.

Ghost wants to change that by keeping it all in your house.

It built Core, a personal AI computer that runs AI models on the device itself.

You hook it up to your email, calendar, and files, plus things like smart lights, speakers, and the ring on your finger tracking your sleep.

Over time, it builds a picture of your life and starts helping without you asking, like reminding you not to forget your keys on the way out.

According to Ghost, none of that data ever leaves your home.

If an assistant is going to know this much about you, it might as well live where you do.

History

AI Reads the Revolution’s Paperwork

The American Revolution happened with many people living through it that didn’t make it into a history book.

But their names are still out there.

They’re scattered across old newspapers and the paperwork veterans filed for their pensions.

And now the Smithsonian and the Library of Congress are digging through all of it with AI.

The project’s called Revolution Crossroads.

It pulls from a newspaper archive of 13M+ pages, focusing on the years around the war.

Plus over 82,000 Revolutionary War pension files that add up to 2M+ pages on their own.

AI is reading every scanned page, layout and all.

It then tags each person, place, date, event, and ship it comes across.

That’s what lets records from different collections link up.

It’s how a silversmith’s newspaper ad from 1774 got matched to a creamer much like one sitting in the National Museum of American History.

The best part is you get to search all of it yourself, for free.

Our take: History mostly gets written from the records someone took the time to read. AI makes the rest of the pile searchable. And that gives people like that silversmith a real shot at being found.

Prompt of the Day

Build My Memory Palace

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

That’s all for Tuesday!

Thanks for reading,

— There’s An AI For That