benstein

I rebuilt isitchristmas.com using 484 subagents and 16 million tokens to learn how Claude's dynamic workflows work. Oh, and today is not Christmas.

I really appreciate how much thought you gave this. Interesting to me that biggest tells came out of the attempts at humor and metaphor.

The title was intentionally ironic :-)

LOL fork

+1000. "Human coders are still better than LLMs" is a hot take. "Antirez is still better than LLMs" is axiomatic ;-)

I made my own webmail client in 2001. The killer feature was a Heart button that one-click opened a new mail addressed to my then-girlfriend-now-wife. I've never seen that feature replicated then or now. And since it was hard-coded to my girlfriend, it was clearly (hopefully!) not broadly applicable.

"Special Olympics T-Ball Stand Pitches Perfect Game" is the greatest headline ever written

The different prompting strategies needed to improve results for different models is fascinating. I usually tell ChatGPT the role it should play to get better results e.g. "You are an expert in distributed systems". The same approach with Gemini returned "as a large language model constantly learning, I wouldn't call myself an expert."

I wrote the article and I think it's funny

Why hate on that game? It had both great puzzles (challenging yet approachable) AND was pushing the envelope of CD-ROM as a new technology, specifically the use of video + celebrities.