stuff tagged with "claude"
vibes is live
Taking inspiration from my buddy Paul's talk, I decided this week that I needed to commit to finishing my patio project.
For those who forgot, I started redoing my patio last summer, but sort of lost steam + got bad anxiety about doing things to my house, so it's sat dormant. A giant sand pit, a daily reminder that I can't finish anything.
The last couple of nights, I've pulled up a patio chair in the sand and imagined what this place can look like. I mean, the hard stuff is pretty much done. I have all the pavers. The retaining wall is complete. Now I just need to clean the sand again, level it, and set the pavers.
I was having a hard time imagining what a good paver layout would look like since I have two different colors, so I vibe coded a little tool to help me visualize what different patterns might look like.
A few weeks ago, my friend Kelly sent out a survey collecting data points for her Minnebar talk. One question on this survey was: "Think of a piece of code you've written that you're proud of. In a sentence or two, what made it good?"
After a lot of reflection, here was my response:
In middle school, I learned how to write TI-BASIC, and I used it to program small apps to help me with math homework. Simple apps that helped me remember things like the Pythagorean theorem.
What made these apps good was sharing them with my friends and seeing them use them for their own homework and tests. I've heard some even used my Pythagorean theorem app in college.
To me, "good code" is mostly unimportant. What's important is unblocking others from being stuck on the semantics (e.g. memorizing a mathematical equation), allowing them to focus their energy on solving a bigger problem.
I was thinking about my Pythagorean theorem app this evening while staring at my patio, and it dawned on me that I have a device in my pocket at all times which Claude claims is somewhere between 200,000Ă— and over 1,000,000,000Ă— more powerful than that humble TI-83+ Silver Edition.
There are a lot of people who have a lot of opinions about artificial intelligence. I find validity in all of them: it's bad for the environment, it's being crammed down our throats, it's trained on stolen data, using these tools fly in the face of hard-earned engineering skills, etc. etc.
Me? My general stance on all technology is "what can this thing do?"
I want to make things. Useful things. Cool things. Fun things.
You know what's fun?
Having an idea for a single page app at 7pm, then hanging out with your family watching Survivor 50 until about 9:45pm, then jumping into a few Claude Code sessions to whip together the idea in about two hours.
If you visit junk.timbornholdt.com/vibes, you'll see a pretty weird looking landing page of random pages.
I have four tools on there right now:
- A visualizer for my patio (better seen on desktop)
- A syllabus guide to the lo-fi genre
- A page designed to encourage me to try new-to-me PS2 games
- Based off albums that I was into last month (need to update now), a page that gives me more detailed listening notes when exploring similar albums
The coolest part? This is all tied into Github Actions. If I want to make another page like this, I load Claude on my iPhone, tell it what app I want it to build, then let it commit that code to my vibes repo, which automatically gets rsyncd to my junk VPS.
I now have the ability to start out on a walk, come up with a random tool I need built, tell Claude to Ralph Wiggum loop itself until it accomplishes the vision, then see the results for myself right on my phone.
It took me two hours to make this happen. Most of it was spent deploying and securing a brand new VPS.
The future is kinda dope.
I hope all of this will help me finish that damn patio.
(PS: Look at how terrible that SVG logo is that it generated lmao I love it so much)
Why Anthropic’s Claude still hasn’t beaten Pokémon
đź”— a linked post to
arstechnica.com »
—
originally shared here on
In some sense, it’s impressive that Claude can play Pokémon with any facility at all. When developing AI systems that find dominant strategies in games like Go and Dota 2, engineers generally start their algorithms off with deep knowledge of a game’s rules and/or basic strategies, as well as a reward function to guide them toward better performance. For Claude Plays Pokémon, though, project developer and Anthropic employee David Hershey says he started with an unmodified, generalized Claude model that wasn’t specifically trained or tuned to play Pokémon games in any way.
“This is purely the various other things that [Claude] understands about the world being used to point at video games,” Hershey told Ars. “So it has a sense of a Pokémon. If you go to claude.ai and ask about Pokémon, it knows what Pokémon is based on what it's read… If you ask, it'll tell you there's eight gym badges, it'll tell you the first one is Brock… it knows the broad structure.”
This is the camp I’m in with AI. Is it super human? Obviously not in this specific instance, but still, undeniably impressive that a large language model is able to get as far as it can.
