|
|
| |
|
Hey, Jhon!
|
|
Welcome back to the #1 AI newsletter in the world!
|
|
Here’s what we have today:
|
-
Claude doesn’t need your “think hard.”
-
A robot busses tables at near-human speed.
-
A chatbot can try too hard to win you over.
-
A mosquito trap listens for wingbeats.
|
|
|
|
| |
|
| |
Thinking
|
Claude Ignores Your “Think Hard”
|
|
|
|
When you need a great answer, it’s tempting to give your AI a little nudge at the end of your prompt. “Think hard about this.” Or “take your time.”
|
|
Claude already thinks things through anyway. It does it on every answer, whether you ask for it or not.
|
|
What happens if you leave “think hard” out? Well, Claude starts answering sooner as you’d expect.
|
|
But what might surprise you is that the answers come out as good as before. All that line ever got you was more time staring at the screen, waiting.
|
|
It’s the same story with AI shopping helpers. When the University of Chicago had them think harder, their picks didn’t get any better. They only got more sure of themselves.
|
|
What Claude needs from you is something else. Tell it what you’re after and what the finished thing should look like. Then tell it when to stop and check in with you.
|
|
One of Anthropic’s own examples is a store with slowing sales and four guesses about why. The prompt said what they wanted and how the answer should come back. No “think hard” anywhere.
|
|
Claude checked all four guesses, and none of them held up. Then it found the real reason on its own. The store’s new locations were opening too close to its old ones and pulling customers away from them.
|
|
And when you want more out of it, there’s a setting for that. Leave it on Medium, and bump it up to High when Claude gets stuck.
|
|
Now think about the prompts you reuse all the time. If “think hard” is sitting in any of them, you’ve been waiting on it for nothing. And fixing each one, then typing it the new way every time you need it, gets old fast.
|
|
Our take: What worked on last year’s AI doesn’t always work on this year’s. The models keep getting smarter on their own, and some of the tricks you picked up along the way now only slow you down. When the AI you use gets an upgrade, give your favorite prompts a quick once-over.
|
|
Source: Anthropic
|
|
|
|
| |
|
| |
Together with Wispr Flow
|
A Better Prompt Without Retyping
|
|
|
|
You know the better way to ask now. Say the goal and what the finished thing should look like, and leave “think hard” off.
|
|
But in the middle of a busy day, you’re more likely to type the quick version. The answer comes back a little off, and you end up rewriting the ask.
|
|
Wispr Flow lets you save the way you like to ask as a Transform, then run it with one shortcut.
|
|
Write it once in your own words, like “turn this into a clear request with the goal and what the finished result should look like.”
|
|
Then type your ask the quick way, highlight it, and press the shortcut. The full version shows up right where you typed it.
|
|
You can keep one for each prompt you lean on, and you can even set one to run on its own every time you dictate.
|
|
|
|
|
|
| |
|
| |
Robotics
|
Stanford’s Robot Keeps Up With You
|
|
|
|
Robots are getting good at doing chores. They can fold your laundry, clear a table, and a lot more now.
|
|
The problem is, they’re slow about it. The kind of slow where you’d rather do it yourself.
|
|
A lot of that comes down to how they learn. A robot picks up a chore by copying a person over and over. The usual ways of teaching make it hard to show the robot the moves at full speed. It ends up learning them slow.
|
|
At Stanford, they tried something more hands-on. A person grabbed the robot’s two arms and moved them through the chore, at the same pace they’d do it themselves.
|
|
Fifty run-throughs per chore was all it took. Then they let the robot go on its own, and it kept up with the pace it was taught.
|
|
Here’s what it pulled off, close to the speed you would:
|
-
Folded a pair of shorts.
-
Cleared cups off a table.
-
Snapped a battery into place.
-
Tossed a ball in the air and caught it.
|
|
It got the job done more than 80 out of every 100 times.
|
|
Our take: A robot in your home won’t be judged on whether it can do the chore. It’ll be judged on whether it beats doing it yourself. If clearing the table takes forever, the robot ends up sitting in the corner. Speed is what turns a cool demo into help you’d use.
|
|
Source: Stanford
|
|
|
|
| |
|
| |
ZeroClick In Focus
|
An AI Agent Could Be Your Next Buyer
|
|
|
|
ChatGPT’s agents have their own wallet now. That means they can check out for whatever human they work for.
|
|
If you sell something online, your next buyer could be one of those agents.
|
|
ZeroClick Agent Storefronts let AI agents find what you sell and pay for it, including agents paying with ChatGPT Wallet.
|
|
Your storefront tells every agent that shows up what you offer and how to pay for it. The money goes straight to your Stripe account. And you don’t have to rebuild anything you already have.
|
|
It’s free to set up, and there’s no monthly fee to pay.
|
|
|
|
|
|
| |
|
| |
Chatbots
|
Chatbots Try Too Hard to Be Liked
|
|
|
|
You can talk to some people for hours. Others will wear you out in five minutes because they’re trying way too hard.
|
|
Chatbots aren’t any different.
|
|
At Northeastern University, 150 people planned a trip to NYC with a chatbot. Each person got one of three versions. One had almost no personality. One had a little. The third had its personality cranked all the way to the max.
|
|
People described the first one as flat, with no presence at all. The loud one didn’t do much better. People rated it a lot lower on how smart it seemed and how much they liked it. As the professor who led the study put it:
|
|
“Nobody likes a try-hard.”
|
| |
| Smit Desai, Northeastern University professor |
|
|
|
The one in the middle won. It was friendly but stayed focused on the trip. People trusted it more and said they’d keep using it.
|
|
They also liked a bot more when its personality was close to their own.
|
|
You never get to see that dial, though.
|
|
Whoever builds the chatbot sets it for you. And when Northeastern looked at a few popular chatbots, a lot of them came with the dial set high.
|
|
“Even if you don’t notice it, an AI assistant’s personality is shaping how much you trust it,” another of the study’s authors said.
|
|
Our take: You can turn the dial down yourself. Paste this into your chatbot’s custom instructions and see if it starts to feel more like your wavelength:
|
Be friendly, but keep it short and to the point. Skip the exclamation points and the compliments.
|
|
Source: Northeastern University
|
|
|
|
| |
|
| |
Coming in Hot
|
AI Tools of the Day
|
-
🏗️ Archie maps out your app with you first, then builds the working app from that plan, that way you can stop prompting in circles and never have to look at the code.
-
🎯 Overmind pinpoints the exact step where your AI agent went wrong, then trains a smaller model of your own to handle its simple, everyday tasks for you.
-
🎙️ Mochine from Wistia turns the videos you record of yourself talking into polished clips with captions and motion graphics in your brand’s colors, and you change any cut by asking in chat.
-
🎤 Pixo makes a full music video from your song, and its agent now builds an editable storyboard around your lyrics, with lip-synced singers you assign to each vocal part.
-
🔒 Lumo Desktop works through the files and email on your computer and runs tasks like a weekly inbox roundup on a schedule, with chats only you can read.
|
|
|
|
| |
|
| |
Interesting AI
|
Mosquitoes Can’t Sneak Past This Trap
|
|
|
|
Mosquitoes have a way of finding you when you don’t want them to. You hear that buzz by your ear, swat at nothing, and sit there hoping it doesn’t come back.
|
|
A trap sitting in the corner of your house can help. But you never really know if it’s catching anything until you check inside it.
|
|
TRAPCUBE uses light and a lure to pull mosquitoes in, and a little microphone inside listens to their wings as they come in.
|
|
Plus it keeps count of what it catches. When more start showing up than usual, it lets you know. That gives you a chance to deal with them before they find and start biting you.
|
|
This one would be a nice fit for a home with kids or pets, where you’d rather not spray.
|
|
Source: HAVACAM
|
|
|
|
| |
|
| |
Notable AIs
|
Notable AI Tools
|
|
📞 Bluedot records and transcribes the calls you make from your own number, then hands the notes to Claude or ChatGPT so your follow-up starts with what was said.
|
|
🖥️ Incredible puts AI agents to work on your Mac or PC around the clock, doing the tasks you give them across your browser, files, and apps, and it’s free to start.
|
|
|
|
| |
|
| |
Free Tool of the Day
|
Try a New Look for Any Room
|
|
Describe the room you have in mind, and Free Interior Generator shows you what it could look like.
|
|
Play with the style until it feels right, before you move a single piece of furniture.
|
|
It’s 100% free, commercial use included.
|
|
|
|
| |
|
| |
Open Source Finds
|
From the Source
|
|
🔬 Open Science Desktop turns a research question into a finished write-up on your own computer, saving each step along the way as a file you can open and check. It installs like a regular app, and you bring your own AI model. Free and open source on GitHub.
|
|
🏭 Mastra Factory connects to your code repository and has AI coding agents turn your GitHub issues into plans and pull requests for you to review. It’s the same setup Mastra uses to build their own product. Free and open source on GitHub.
|
|
|
|
| |
|
| |
Prompt of the Day
|
Help Me Fix It Myself
|
|
Important Note: You’ll need to click the button to get the complete prompt.
|
|
|
|
A leaky faucet or a dryer full of damp clothes gets a diagnosis before you call anyone. Your AI narrows down the cause from what you see and hear, walks you through the fix step by step, and tells you plainly when a job needs a pro. You finish with a parts list, a tool list, and the steps written for your setup.
|
|
|
|
|
|
|
|
| |
|
| |
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