zlacker

>>Josely+(OP)
This is a bit off topic to the actual article, but I see a lot of top ranking comments complaining that ChatGPT has become lazy at coding. I wanted to make two observations:

1. Yes, GPT-4 Turbo is quantitatively getting lazier at coding. I benchmarked the last 2 updates to GPT-4 Turbo, and it got lazier each time.

2. For coding, asking GPT-4 Turbo to emit code changes as unified diffs causes a 3X reduction in lazy coding.

Here are some articles that discuss these topics in much more detail.

https://aider.chat/docs/unified-diffs.html

https://aider.chat/docs/benchmarks-0125.html

>>anothe+Pf
I have not noticed any reduction in laziness with later generations, although I don't use ChatGPT in the same way that Aider does. I've had a lot of luck with using a chain-of-thought-style system prompt to get it to produce results. Here are a few cherry-picked conversations where I feel like it does a good job (including the system prompt). A common theme in the system prompts is that I say that this is an "expert-to-expert" conversation, which I found tends to make it include less generic explanatory content and be more willing to dive into the details.

- System prompt 1: https://sharegpt.com/c/osmngsQ

- System prompt 2: https://sharegpt.com/c/9jAIqHM

- System prompt 3: https://sharegpt.com/c/cTIqAil Note: I had to nudge ChatGPT on this one.

All of this is anecdotal, but perhaps this style of prompting would be useful to benchmark.