lukasco

5 ms·

1. 31.▲
   by lukasco 3mo ago

   Totally agree. And really, we use the llms to be the universal layer, not the harness. It’s already nigh on impossible to eval a harness with multiple turns (at least as far as I’ve seen), so multiply by specific llm prompt…)
2. 32.▲
   by lukasco 3mo ago

   Fun, seems to be working. Do you have levels? All the games I played seemed to have a static level.
3. 33.▲
   by lukasco 3mo ago

   This is the big question! Is it really sensible to distribute locally running software at all? Maybe this leads to the locked down Mac App Store (and Windows I'm sure)? Apple's been wanting to lock down the Mac for ages, and this
4. 34.▲
   by lukasco 3mo ago

   It sounds like harnesses might have to start to have model by model system prompts, though retrying works, I guess. It reminds me of the ancient times when browsers all read HTML and CSS differently, and differently on different devices. I
5. 35.▲
   by lukasco 3mo ago

   It's not YOLO, but auto mode in Claude Code does reduce the amount you have to approve significantly. And frankly, without it, progress is constantly interrupted by permission requests. It's all I use. Don't even really switc
6. 36.▲
   by lukasco 3mo ago

   This is the big thing I'm struggling with: Code reviews are still a critical step in most workflows. Though seems like everyone uses them for a different purpose: extra pair of eyes to meet a regulation/security, style police, and