|
|
| |
Hey, Jhon! |
Welcome back to the #1 AI newsletter in the world! |
Here’s what we have today: |
OpenAI says its AI broke a 27-year rule. An old AI reads skin spots better than chatbots. The words people use lean toward safety. Anthropic officially files to go public.
|
|
|
|
| |
|
| |
Discovery |
Experts Check AI’s Puzzle Shortcut |
|
|
There’s a rule in computer science called the Exponential Time Hypothesis. |
It’s one of those behind-the-scenes ideas computer scientists have leaned on for 27 years. |
Now OpenAI says their AI broke it. |
The easiest way to understand it is with a road trip. |
Planning a route through five stops where you hit each one once and never backtrack is no big deal. |
Make it 50 stops, and there are so many possible routes that even the fastest computers would run out of time trying them all. |
The rule says puzzles like that have no real shortcut. |
And it’s built on a classic logic puzzle that’s tied to a whole family of them, including route planning and scheduling. |
It’s never been proven, but researchers trusted it enough to make it the go-to way of saying “this problem won’t get much faster.” |
“Surprisingly, our system has recently proved that the ETH is false,” OpenAI wrote. |
It’d be easy to write that off as hype. |
A proof-checking program called Lean went through the math line by line and signed off on it. |
OpenAI announced it alongside two more results, in geometry and physics, including part of one of the Millennium Prize problems, the famous list of math’s hardest open questions. |
Instead of taking a victory lap, they handed the computer science result to Russell Impagliazzo and Ramamohan Paturi (the two researchers who came up with the rule back in 1999). |
Picture watching an entire field build on your idea for decades, then getting asked to check whether a machine broke it. |
Ryan Williams of MIT is checking with them, and the physics proof went to a group that includes Nobel Prize winner David Gross, staying under wraps until they finish writing it up. |
It’s not a done deal yet, and OpenAI pulled three of their own AI-written papers on Oct 7 after a sign error turned up. |
But if this one holds, a lot of “this can’t be done faster” results go back on the table. |
Our take: A company grading its own breakthrough isn’t worth much, which is why this setup matters. The people whose rule is on the line have more reason than anyone to hunt for a mistake. If they sign off, that’ll count for more than anything OpenAI could say about it. |
|
|
|
| |
|
| |
Together with Floot |
An App Your Whole Family Can Sign Into |
|
|
Holiday gift lists have a way of falling apart in the group chat. |
Somebody buys the same thing twice, and the list everyone swore they’d keep updated hasn’t been touched since last year. |
Claude or ChatGPT can create a little app for it in a few minutes (a shared list where everyone adds ideas and quietly marks what they’ve already bought). |
Getting it out of the chat and onto everyone’s phones, with their own logins, is where it gets stuck. |
Floot plugs into the AI you already use and turns the app you built in the chat into a real one your family can sign into. |
It sets up the database and the logins in that same conversation, then puts your app live at its own link. |
Connect your developer account, and it’ll ship it to the App Store and Google Play as well. |
Your own AI does the building, which means you’re not paying for a second set of credits on top of the subscription you already have. |
It even works on Claude’s free plan. |
|
|
|
|
|
| |
|
| |
Health |
Chatbots Misread Skin Spots |
|
|
A skin cancer surgeon ran a little experiment on chatbots and moles, and the chatbots didn’t come out of it looking great. |
He pulled 200 skin spots where the answer was already known, since every one had been biopsied, and showed them to two of the most popular chatbots. |
He also gave the same photos to a much older AI, about 10 years old, that was only ever trained on skin. |
On paper, the new chatbots should’ve won that one easily. |
Instead, the old skin AI got the right answer about six times out of ten, and the chatbots got closer to four. |
Worse, they called a few melanomas harmless and told people plenty of harmless moles were cancer. |
In fairness to the chatbots, when he asked the way a worried person would, with one regular photo and “do I need to see a doctor?”, they played it safe. |
They sent people to a doctor for nearly every melanoma, and for most of the harmless spots as well. |
His advice is pretty simple. |
Don’t trust a chatbot that says a spot is fine to ignore, and get anything that’s changing or bleeding looked at by a doctor. |
The catch: It’s a small test, and biopsied spots are harder to call than your average mole. He also says the chatbots might’ve seen these public photos before. |
Our take: Use a chatbot to help you decide whether to book the appointment, never to decide you can skip it. A tool built for one job beat two that do everything, and that’s worth remembering any time an all-purpose AI sounds sure about something specialized. |
|
|
|
| |
|
| |
Coming in Hot |
AI Tools of the Day |
🎬 Clueso writes and voices a product video from your description, then re-records the footage on its own whenever your product changes. 🗺️ Plantripr builds a day-by-day trip with routes and food stops, then lets your friends vote on activities through a shared link. 🖌️ FlowGen turns a quick sentence about the screen you want into a live, editable design, ready to export as React code once it looks right. 🏷️ SellRaze posts one listing across eBay, Facebook, Mercari and more, then keeps every buyer conversation together in one place. 🤖 OpenShell lets your AI agents touch only the files and websites you approve and never shows them your real passwords, free and open source on GitHub.
|
|
|
|
| |
|
| |
Language |
People Talk Safety More Than Danger |
|
|
The same University of Vermont team that built a “happiness meter” for social media went back and poked holes in the idea it was built on. |
For decades, research on how words make people feel has leaned on one main question, and a lot of AI picked up the habit along the way. |
Here’s how the two maps compare. |
The old map asks how good or bad a word feels. |
The new one asks first whether it feels safe or dangerous. |
After going through billions of words, old novels and talk radio included, the team found the new question explains a lot more. |
It covered over 90% of how people read a word’s meaning, against about 72% for the old map. |
And safety wins by a mile. |
People reach for words about feeling safe far more often than words about danger. |
Pretty much everywhere the team looked. |
They think the long-known habit of people sounding more positive than negative is this safety lean showing through. |
Their new tool, built on top of the happiness meter, even tracked how the mood shifts across Les Misérables. |
It matters outside research papers as well, since some of the tools that judge whether a review sounds angry or a post sounds upset still run on the old map. |
The team is careful to say none of this proves people’s brains evolved this way to survive, only that the pattern fits. |
Our take: A thermometer that only reads hot or cold misses a lot about the weather, and mood-reading AI has been working off something close to that. Teaching it to notice whether words feel safe or threatening should help it tell a worried message from an angry one. |
|
|
|
| |
|
| |
Interesting AI |
Your Next Robot Looks Like a Lamp |
|
|
Home robots in movies come with arms and a face on a screen. |
Mori skips all of that. |
It’s a soft, glowing little thing about the size of a table lamp that sits on a shelf and keeps you company. |
In the morning it gives a small tilt and a curious glance to say hello, and when you walk in the door, it greets you in its own way. |
Touch it, and it fills the room with a nature scene in light and sound, picked for your mood, like waking up or winding down after a long day. |
OKA says everything it senses stays on the device, and that it remembers and learns more about you the longer it’s around. |
|
|
|
| |
|
| |
Business |
Anthropic Officially Files for IPO |
|
|
Anthropic is going public. |
They filed the paperwork to list on the Nasdaq under “ANT,” and since a company has to open up their books before anyone can buy their stock, we finally get a real look inside. |
The growth is something else. Revenue in the first half of 2026 came in at about 14 times what it was a year earlier. |
They’re also still losing money, and they have every year since they started. |
That loss looks enormous on paper, but most of it’s an accounting entry from their own IOUs going up in value, not cash leaving the building. |
They’re upfront that growth this fast won’t last, and they say they might need to raise more money sooner than planned. |
A big chunk of what they spend goes to computing power, locked in for years. That includes deals with Google and Microsoft that run into 2033, and they pay whether they use it all or not. |
There’s another one with SpaceXSI’s AI arm, the company behind Grok, to rent Nvidia chips into 2029. |
The bigger surprise is who stays in charge. |
The seven co-founders get one special share that carries a little over half the votes, and a trust with no money in the company picks most of the board. |
People who buy the stock won’t get much say, and the filing spells that out as a risk. |
They’re also setting aside 5% of the company for a fund dedicated to the public good, with more added each time the stock climbs 50%, and the co-founders have each pledged 80% of their own shares to charity. |
Tucked into the list of risks, they say their models are “contributing to the pace of our own research,” right next to a warning that advanced AI could “pose catastrophic or existential risks to humanity.” |
Our take: Going public means answering to shareholders every three months. If the founders’ votes and the trust hold up the way they’re designed to, Claude shouldn’t change much once investors own a piece. If that setup starts to bend, it’ll show up in what Claude gets built to do long before it shows up in the stock price. |
|
|
|
| |
|
| |
Prompt of the Day |
Level Me Up |
Important Note: You’ll need to click the button to get the complete prompt. |
|
|
This prompt turns your real life into a role-playing game character sheet, with a class, a level, six stats, a skill tree and the debuffs slowing you down. You finish a side quest worth real XP before the chat closes. |
|
|
|
|
|
|
|
| |
|
| |
That’s all for Monday! |
|
|
How did we do today?Rate this issue, then tell us what you thought. We value your input and will use it to make TAAFT better for you. |
|
|
|
|
Thanks for reading, |
— There’s An AI For That |
|
|
|
| |
|
|
Postingan