[onewhippedpuppy]

AI is very good with things that generally have strong guardrails and/or rules are are very repeatable. Commands to touch up photos are generally the same, and AI can use literally the entire internet of photos to determine what things “should” look like. Programming is another one, the rules are very well defined and common across a given flavor of code. Reviewing legal documents against the applicable laws, summarizing a document, reading a CT scan, all of these are very well defined tasks with an abundance of data to train the AI/ML model.

It struggles when you inject ambiguity. Fact vs fiction vs propaganda, good sources vs bad, applying basic human rules of right/wrong when determining outcomes, etc. It’s basically a master plagiarizer but struggles to pick the right sources sometimes. It’s also not actually intelligent, just an aggregator. Useful, but not blindly so.