Comment by rvnx
2 days ago
In the tests it more obvious:
You can see here for example: https://github.com/piyushgupta53/go-torrent-client/commit/61...
and some strings coming from crawled resources like: lengthi12345e4 but slightly different tokens (like 25 instead of 35 etc).
Gemini Pro 2.5 even gave me the prompt:
> If you asked me, "Generate Go unit tests for a Bencode decoder function called Decode that takes an io.Reader and returns an interface{} and an error. Cover strings, integers, lists, and dictionaries, including common error cases and nested structures" the output I would strive to produce would look very much like the code you've shown.
> It's a good example of well-written Go tests, and that's the kind of pattern I've learned to recognize and replicate.
and a lot actually matches when you ask from a fresh conversation.
So most likely Cursor + Gemini 2.5 Pro, but I cannot blame, I spend 100% of my time with Claude, and I take ownership of the code.
"TODO: We'll develop the actual functionality as we develop each component" lool
It's hard to say honestly. I don't call any project AI as it's just too hard to tell. I write lots of comments in my code too so it's hard to call anything AI without a person stating they used it.
Claude is decent for sure, but I always say with AI, learn the math before jumping to a calculator.