nostr:npub1f6a33pfyp67y8llhunlhrf855xm47n3fdqymvxfj7yx78c6vqf4scxpnql is GPT-4 significantly better for coding than 3.5? I confess to have used it myself when I'm stuck or want a different angle - certainly easier than trawling stackoverflow - but usually I find enough flaws that I end up sinking hours into the problems. I wonder if the problem is the 2019 data (new lobs and frameworks have moved on) or whether it is something more fundamental.