wilbur_whateley

How did he punish them?

My own experience. I'm working on something complex that's not in the datasets these models were trained on. There I see V4 flash breaking down and hallucinating much more often than GPT/Claude. For normal, common tasks, I also don't see much of a difference.

V4 flash is much worse than any Claude model. If you're doing something simple, it can be a good way to save money though.

Working on Nichess (chess with health points) - https://www.nichess.org/

If you like this, you might find Nichess interesting - https://news.ycombinator.com/item?id=47947041

Adding health points to chess gives you a lot more options for balancing and abilities.

Games from that time had so much soul. I wish I were better at the art side of things.

You're right, I'll add that. You remember them after playing for some time, but it is useful for a new player.

Yes! Special abilities (Queen-Maenad and Pirate-Pirate) deal 20 damage to neighboring pieces. The only difference between 10 HP pieces and 30 HP pieces is that 30 HP pieces can survive that.

AI playing Nichess 1 against itself - https://youtube.com/playlist?list=PLN0TnHkszmgQSnFrvF_ANIzX8...

Claude with Sonnet medium effort just used 100% of my session limit, some extra dollars, thought for 53 minutes, and said:

API Error: Claude's response exceeded the 32000 output token maximum. To configure this behavior, set the CLAUDE_CODE_MAX_OUTPUT_TOKENS environment variable.

Working on https://www.nichess.org/

Nichess is a game like chess, where pieces have special abilities and health points. This allows for much finer balancing and many more variants compared to the original chess. It will take some time, but it will become great eventually.