Most of those look like context issues to me. Repo map (using Tree-sitter, etc) and documentation would already do wonders. Feeding 32-64kTok of context directly into a model like Gemini Pro 2.5 is something that more people should try out in situations like this. Or even 128kTok+.
Are you using a tool other than ChatGPT? If so, check the full prompt that's being sent. It can sometimes kneecap the model.
Tools having slightly unsuitable built in prompts/context sometimes lead to the models saying weird stuff out of the blue, instead of it actually being a 'baked in' behavior of the model itself. Seen this happen for both Gemini 2.5 Pro and o3.
A lot of people make small games just for fun, like any other creative hobby. Similar to: Making music, writing, drawing, 3d modeling, etc.
Actually, you can put all of those together and use them in a game. And the best part is that there's no target market, no KPIs to hit, so you can do anything!
He addressed your point in the paragraph before that. The paragraph from which you quoted was meant to show the difference between your point and the fact that the original research was indeed measuring software engineers.
That's a strange comparison! I guess this is where the ultimate caricature of cool comes from, though: Sex, drugs and rock 'n' roll -> Hedonistic, powerful and adventurous. Crank those up enough, and you end up with a trashed hotel room or maybe a drumstick up someone's butt.
reply