The iPhone 4 was released on June 24, 2010, over 15 years ago. It was a groundbreaking phone, introducing the Retina display that is now built into every modern smartphone. It also was the first iPhone with a front-facing camera, allowing FaceTime video for on a phone. It had a strong CPU built in and shipped with a mature operating system, allowing multitasking and app folders.
It was a great package in a new form factor. It was also the first iPhone to get over 600,000 pre-orders in 24 hours. Most customers saw this phone as far ahead of the competition and a good buy. For many, it was their first iPhone.
I see many similarities to Opus 4. Working with it, you can feel it’s a mature model that has been iterated multiple times. There are no rough edges, it’s just a pleasant model to work with.
Anthropic bet big on Opus 4, training it extensively on coding tasks, and this paid off. In their marketing, coding is the first strength they mention. That’s why we see other companies going crazy. Meta is paying NBA salaries. Google just did an acqui-hire of Windsurf for 2.4 billion USD. They realize they don’t have such a model and due to the training cycle these models take, they won’t for a foreseeable time. The American tech landscape is defined by a winner-takes-it-all mentality, and for the first time in the AI hype, one company has a substantial advantage.
One concrete example from my last week: I am trying to get fast local LLM inference on my AMD Ryzen AI Max mini computer. It’s a great machine, more or less an M4 Pro with 128GB of RAM. There’s just one catch: The driver situation from AMD is horrible. I read a blog post, detailing what it takes to get it running. It involved updating to the latest Linux kernel and installing a nightly tarball of an AMD library. Because this machine doesn’t hold private data and I can easily set it up again, I decided to give Opus 4 a chance to do this for me.
Admittedly, it took a while, and it needed my help. It couldn’t reboot the machine by itself and continue working. It tried a few things that didn’t work. But ultimately it succeeded! I now have GPU support for running local LLMs on this machine. I just gave Opus 4 a starting point with a blog post containing the right information.
I wouldn’t trust any other model for this, and I don’t plan on doing a test series for this at the moment. Opus 4 is in a different league than all other models. That’s why it’s the iPhone 4 of the AI era.
Funny sidenote: No groundbreaking product is perfect. The iPhone 4 had a severe antenna issue leading to Antennagate. I remember ordering these band-aids with the funniest marketing claim: “Apple made a boo-boo. Make it all better”. They solved the problem and looked hilarious at the same time.
What I Learned this Week
This talk goes deep and get's so much right. Comparing our constitution/Grundgesetz to a spec is a such a powerful thought, where this could all go. VIDEO
147k LOC. And you are telling me, vibecoding is just for pet projects? 98% written with Claude Code. LINK
Long piece, worth to skim over. So few people have a lot of experience working with agentic coding, even fewer share it in such a way. LINK
What to Print this Week
This newsletter started out on 3D printing. If you haven't had any contact with it, you should, it's great! Here's the most interesting and funniest projects I saw last week.
Watch the video. Still amazing what people can build with the help of 3D printing.
|
|
Port Nexus: Modular Shipping Port
|
Great name, great idea, great execution: Conquering walls near you!
|
|
Shelfinity - Modular Shelf System
|
Well designed.
|
|
Modular desktop storage box
|
|
|
Hi 👋, I'm Stefan!
This is my weekly newsletter about new technology hypes in general and AI in specific. Feel free to forward this mail to people who should read it. If this mail was forwarded to you, please subscribe here.
|
|
