|
|
| |
|
Hey, Jhon!
|
|
Welcome back to the #1 AI newsletter in the world!
|
|
Here’s what we have today:
|
-
Claude finds your money leaks.
-
One prompt built a playable tennis club.
-
A red heart changes how AI writes code.
-
AI agents build on their own mistakes.
|
|
|
|
| |
|
| |
Money
|
Claude Finds Your Money Leaks
|
|
|
|
The end of the month has a way of sneaking up on your bank account.
|
|
You open the app, scroll through a few weeks of charges, and still can’t say where it all went.
|
|
Claude can now go through all of it for you.
|
|
Anthropic added Personal Finance, a new section in Claude’s sidebar that connects to your bank and card accounts.
|
|
Ask for things in your own words.
|
|
|
|
It adds up what you spent across the accounts you connect and lines it up against the months before.
|
|
It also goes looking for the charges that slip past you, like a double charge or a bill that crept up over the year.
|
|
Ask what’s left before payday and it lists the payments still coming out of checking.
|
|
It can build you a chart or a calendar of due bills, and it’ll explain a term like APR along with what it’s cost you.
|
|
Hooking up your bank doesn’t mean handing over your password. You sign in to your bank through Plaid, a service banking apps use to link accounts, and Claude never sees it.
|
|
It can’t move any money either. Access is read-only, and it can’t pay or invest anything for you.
|
|
It’s in beta on Claude’s paid plans, and it can look back over up to two years of transactions.
|
|
Our take: The first answer is the least useful part. A category total won’t tell you a streaming plan quietly went up twice, but Claude can point to the exact charges behind any number it gives you. Ask for those before you cancel or move anything, because the AI can still get a figure wrong.
|
|
Source: Claude
|
|
|
|
| |
|
| |
Together with Multiverse Computing
|
Run Your Favorite Models Cheaper
|
|
|
|
You built a little chatbot to answer questions on your website.
|
|
A month later, more people are using it, and the bill grows with every question they ask.
|
|
A big part of that bill comes down to who runs the model for you.
|
|
Multiverse Computing’s CompactifAI API runs popular open-source models like GLM and Qwen for less than other providers charge for the same models.
|
|
They’ve also built their own compressed models, which stay close to frontier quality for a fraction of the cost.
|
|
And those run up to twice as fast. Moving your chatbot over takes a few minutes.
|
|
It works with anything built for OpenAI’s API. Point your code at their address with a new key, then pick a model.
|
|
Their official node plugs it into your workflows without code if you’re building in n8n instead.
|
|
And if your data has to stay in Europe, their EU endpoint keeps it there.
|
|
|
|
|
|
| |
|
| |
Vibe Coding
|
One Prompt Builds a Tennis Club
|
|
|
|
One hour and 40 minutes is how long Claude spent building an entire tennis club from a single prompt.
|
|
The person behind it wrote out everything they wanted in one long message.
|
|
A pixel-art stadium on a floating island, and a full match that ends in a trophy ceremony with confetti.
|
|
Claude came back with all of it in one web page. Two computer-controlled players go through full matches with real ball physics while you watch from above. You can switch the court between grass, clay, hard and carpet. There’s a full day-night cycle as well, and every racket hit and crowd cheer is made right in your browser.
|
|
The prompt even asked for a coach mode where you call the shots, like rushing the net or hitting a lob.
|
|
And the whole club fits in a single file.
|
|
Our take: The model gets the credit here, but the prompt did most of the work. It reads like a contractor’s spec sheet, with exact camera angles and even how often each serve should go in. Hand a builder a sketch on a napkin and you get their best guess. Hand them plans with every measurement and you get what you pictured. Before your next big ask, write down what “done” looks like in that much detail.
|
|
Source: Vib3Coded on X
|
|
|
|
| |
|
| |
Maket In Focus
|
Your Dream House Finally Gets a Floor Plan
|
|
|
|
You’ve walked through enough open houses to know what you want. A bigger kitchen, and an office far from the bedrooms.
|
|
But every floor plan you find online comes close and never fits.
|
|
Maket draws yours. Describe the rooms you need and the size you’re after, and you get several floor plans to compare in a few minutes.
|
|
If you don’t like where the stairs landed, simply describe the change in your own words and the stairs move. Then see the place furnished in 3D, in any style you like.
|
|
|
|
|
|
| |
|
| |
AI Quirks
|
Hearts Change How AI Writes Code
|
|
|
|
A red heart has nothing to do with writing code.
|
|
Google DeepMind showed AI models a small picture of one anyway, right before asking them to solve a simple coding problem.
|
|
The problem had two right answers. One is a step-by-step loop. The other is recursion, where a piece of code keeps calling itself until the job is done.
|
|
Without the heart, two of the biggest AI models never once went with recursion in 20 tries.
|
|
With the heart in front of them, both picked recursion all 20 times.
|
|
Describing the heart in words instead of showing it barely moved them. Only the picture worked.
|
|
Triangles and circles pushed one of the models toward recursion too.
|
|
And the models don’t seem to know it’s happening. When one model laid out its thinking, it never mentioned the picture at all.
|
|
Another model called the shape irrelevant in most of the answers it had changed.
|
|
Source: arXiv
|
|
|
|
| |
|
| |
Coming in Hot
|
AI Tools of the Day
|
-
🎵 freebeat’s Song Finder names the song stuck in your head when you hum it or paste a video link, then turns the match into a music video right in your browser.
-
🌐 WebsitePublisher AI builds and hosts your site or online store from a chat with ChatGPT or Claude, and remembers the project across chats so you pick up where you left off.
-
🧭 Ayven hands approved tickets to your own AI agents through Jira or YouTrack, and nothing moves to the next stage until the person who approved it says yes.
-
🎯 Alsona ranks accounts by how well they fit the customers you want, and when they reply on LinkedIn or email, AI drafts your answer in one shared inbox.
-
🔎 Programmable researches companies that match one line about who you sell to and drafts outreach to the right person, then holds every message until you give the go-ahead.
|
|
|
|
| |
|
| |
Notable AIs
|
Notable AI Tools
|
|
👂 Athena can now hear you, and answer. Talk to it in Ask, cut in any time, and what you say becomes notes, tasks and goals that show up in your next morning read. It works in 24 languages and never stores your audio, and it’s free to use.
|
|
|
|
| |
|
| |
By TAAFT
|
AI Speeds Up the Fight Against Aging
|
|
|
|
For a century, fighting aging worked one slow guess at a time.
|
|
One scientist spent years testing genes by hand to find the few that wind a cell back to its earliest state.
|
|
AI is cutting those years down.
|
|
A model from OpenAI rewrote two of those genes, and in lab tests its versions worked far better than the originals.
|
|
Another AI designed a gene editor unlike anything found in nature, and it edited human DNA with far fewer mistakes than the classic tool.
|
|
And three of the original genes have now gone into people, in a shot for patients going blind from glaucoma. Two of the three can now see more than before.
|
|
None of this adds a day to your life yet.
|
|
But every step is moving faster than it ever has.
|
|
We break down how AI is changing the race against aging, and what it means for you, in our latest video.
|
|
|
|
| |
|
| |
Free Tool of the Day
|
Turn Anything Into Calendar Events
|
|
Dates come at you from everywhere, like a school flyer or an email with the details buried halfway down. Adding each one by hand is the part that never happens.
|
|
Agenda Hero turns any of it into calendar events. Paste the text or drop in a photo or PDF, free in your browser.
|
|
|
|
| |
|
| |
AI Agents
|
Coding Agents Build on Their Own Mistakes
|
|
|
|
What happens when you hand an AI coding agent 20 requests in a row instead of one?
|
|
Each request builds on the last, the way real projects do. Add a login page, then a settings screen that depends on it, and so on.
|
|
A new benchmark called BOBA put ten coding agents through 280 chains like that, pulled from 140 real open-source projects.
|
|
Handled one at a time, the agents got 95% of requests right.
|
|
Chained together, they got 51% right.
|
|
The code itself wasn’t the problem. The agents call a step done once their own tests pass, even when those tests miss a mistake. The next request then gets built on top of the broken one, and requests built on broken code fail about twice as often.
|
|
When the agents were asked to review code and say whether it worked, they were right only 53% of the time.
|
|
Our take: You might read this and decide agents can’t handle long jobs. The fix in the paper is a lot smaller than that. One outside “that’s wrong” after a failed step, plus one round of repairs, lifted chains to 59%, and a fixed mistake stopped hurting the steps after it. Check work you get back after each big step, not only at the end, and run the app yourself rather than trusting the agent’s own tests.
|
|
Source: arXiv
|
|
|
|
| |
|
| |
Open Source Finds
|
From the Source
|
|
📱 Off Grid AI runs a private Al on your phone or Mac that chats and makes images with the Wi-Fi off and no account, free and open source on GitHub.
|
|
🛡️ Stroq scans your coding agent’s past sessions for passwords and keys sitting in them, then flags risky commands in the next session, all on your own machine, free and open source on GitHub.
|
|
|
|
| |
|
| |
Prompt of the Day
|
Spec Out Your Next Build
|
|
Important Note: You’ll need to click the button to get the complete prompt.
|
|
|
|
This prompt turns a rough idea for an app or a small tool into a detailed build brief an AI builder follows step by step. The result is a ready-to-paste build prompt plus a checklist for testing each step before moving on.
|
|
|
|
|
|
|
|
| |
|
| |
That’s all for Thursday!
|
|
|
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
|
|
|
|
| |
|
|
0 komentar:
Posting Komentar