How accurate must AI code and content generation be?

The goal of content generation is to create enough value for people to find it interesting and start sharing across the web. LLMs are great at providing historical datasets around many topics, but not as good as creating new innovative solutions, yet. That being said, I've seen approaches where AI content is used at 100% rate, and SEO and other metrics kick in, hence that's useful on its own. It all depends what's your goal and metrics are; needles to say, AI-generated content has gotten a whole lot better than 1 year go, and it's only going to be improving, so you definitely want to be using it at some portion.

Depends on use case - 99.9% for medical/financial, 80-90% fine for content generation with human review.

Using some percentage of AI-generated content is fine as long as the overall quality and user experience are good. I think benchmarking SEO metrics is key to finding the right balance. AI accuracy will keep improving over time so it's worth experimenting with AI tools now to stay ahead of the curve. I've heard good things about Claude AI being better than GPT-3 for content generation recently.

It really depends on your specific use case and goals. If you just need AI-generated content to kickstart SEO and get some initial traction, even imperfect AI output could work. The algorithms keep getting better so you can likely use a higher % of AI content over time. But for mission-critical or highly innovative content, AI isn't quite there yet to fully replace human writers and subject matter experts. I'd suggest using AI as an assistant to help generate ideas and first drafts, then have a human refine it. Tools like GPT-4 and Claude are pushing the boundaries of what's possible.