ChatGPT experiments: Snake

For those of you who were born in the 80s have probably played snake on a feature phone from that era, perhaps a Nokia 1100 or 3110, those indestructible bricks that you carried around and used for calling, texting and playing snake mostly.

So I have done a few experiments with ChatGPT 3.5, and today I decided to get ChatGPT to write the game snake for me in the language Go. This is a note on my experience with it. See the chat log here.

Overall my conclusion:

1. It created the initial code pretty soon, but the first try was terrible and the cursor had no control at least on my Mac, it wasn’t until later that it did figure there might be changes for Mac.
2. I had to poke and prod quite a bit for it to produce anything that seemed to work, it also threw out some code samples with errors (unrecognized variables) which was surprising, maybe it’s picking up code samples from elsewhere.
3. My last attempt led to a runtime error, interesting, I didn’t dig into it more today but might do so tomorrow.
4. This would be much faster if you knew how to code and were just looking for help, not if you have no idea what the code that the system is generating is really. So it does seem like it might be a good copilot, but not a good pilot. Sort of reminds of the Tesla autopilot option, you still need to be ready to take control.