> allowing it to use debuggers, benchmarks, and sanitizers, allowing it to write its own tools, and then put modules together That’s a huuuge if whether it can

> allowing it to use debuggers, benchmarks, and sanitizers, allowing it to write its own tools, and then put modules together That’s a huuuge if whether it can meaningfully reason about such. GPT-4 has improved a lot over 3, but I really wouldn’t call its capabilities reasoning at all. We often under-appreciate human intelligence