Some working notes on using GPT-4 to code (and LLM’s generally)
- GPT-4 is definitely the move over GPT-3.5
- It’s pretty clear when GPT loses the context window - sometimes it just makes more sense to paste the code into a new chat window to refresh.
- Small contained steps are the way to approach the problem. Doing a ton of modeling / data stuff, the way seems to be “Looking for a dataframe that looks like this” over “I want a program that deals with this whole thing in one shot.”
- Related: I haven’t really dealt with trying to get any data. Because the training log is capped in late 2021, would rather spend the development cycles with code rather than surfing the web.
- The browsing plugin is fine. Not great, and there definitely is some manipulation needed to get it to actually surf the web. “Please Browse the Web” is an A+ prompt technique to do so
- The other plugins definitely appear to be in Beta. I tried Notable and it never made a notebook. Need to continue to test out instacart and expedia.
←
Analytics Motto
Installing nfl_data_py on Mac
→