I'm taking a break from Twitter. Current writing is on

My tech talk in case you missed it

RT @chia_project: Don't forget! This Thursday, July 10 @ 12:30 PT we will be live with @bramcohen with another Tech Talk. Join us for the l….

RT @chia_project: It's time for another Tech Talk with @bramcohen! Join us next week, July 10th @ 12:30pm PT. We'll chat with Bram about al….

RT @steppsr: 7/7 @steppsr - Krunk & Poker Prototypes -- Bram Cohen’s HTML prototypes for Krunk, Space Poker, and California Poker? Steve St….

In case you're wondering, I've been vibe coding front end HTML for Chia games. Partially this is to learn about vibe coding, but mostly it's because I'd like to never look at HTML or Javascript ever again.

AIs are sycophantic blithering idiots

RT @chia_project: Today, we’re excited to launch the early access release for the Chia Cloud Wallet!. Your Keys, Your Device, Uncompromisin….

Claude has self awareness about its lack of self-awareness

Chialisp Programming

A guide to what objects look like when they rotate for AIs

Advice on vibe coding

Some improvements on my memory game

Here it is with code cleanups, the randomness unbiased, and cosmetic improvements

It also supports keyboard navigation, using space/enter and 1-4 for selections.

I'd been meaning to get around to implementing this for years and just now did it in a few hours of vibe coding with the new Claude. The HTML might be an unholy mess, I literally haven't looked at it and would be curious to hear an opinion from experts.

So-called 'Brain Training Games' suck. They train you to do things which are unnatural and weird. Your instincts are usually correct and training to ignore them is bad. Thinking with a stack is very common and important. Here's something to train that:

Claude 4 by contrast writes code to draw a nice tic tac toe board but can't see where I'm clicking on it so I have to describe. It then proceeds to play accurately until the very end where it fails to block a win and loses, then argues with itself about whether it's just lost.

I just beat it playing tic-tac-toe by opening on 8 then playing 3. It's the same 'out of opening book' trick which usually works when playing against little kids with some experience. It took some prodding for it to notice I'd won.

I asked it 'Can you write a Python program to make the best possible chinese poker hands from a set of 13 cards and run it on that example?' and after a deep think it gave the right answer haha.

I just asked ChatGPT to make a randomly generated Chinese Poker hand and make the best hand it could out of it. It made ♠️3♦️9♥️J♣️4♠️10♦️2♥️5♣️Q♠️ 6♣️7♦️7♥️A♠️K but even after repeated nudging failed to realize it could make both straights at once.