This was solid. Really appreciated this. You're one of the few people writing about these tools without sliding into hype or fear. I’ve used AI for research too. Sociological analysis, long-form, productivity and I’ve landed in the same place: it’s useful, but only if you know what it’s actually doing.

The hallucination issue doesn’t go away. It never will. It just changes shape depending on whether you're giving it real data or asking it to go find some. Deterministic inputs - summaries, formatting, rephrasing known content, are cleaner. But once you step into speculative tasks, you’re back in probabilistic mode. More insight but more prone to hallucination. Could you restrict AI to only objective tasks? Sure. But then, you might as well just use a spreadsheet. The real power in the AI tool is accepting but also mitigating the risk when speculating. And not using output as conclusion.

So yes, AI tools are assistants, not oracles. They can accelerate the stages, gathering, organizing, framing but the thinking? Still yours.

Appreciated how you framed it: assistant, not answer key.