[jqpabc123]

and synthesizing the results into an answer.

*An answer* but not necessarily a correct or accurate answer --- and therein lies the problem.

[verdverm]

Humans do the same thing, how is the problem different?

With Ai, I can have multiple agents review the synthesis. When given original source text, they are now quite good at piecing it together with citations. You still have to review as a human, just like when evaluating all the source material when using trad search and clicking a bunch of links. The main advantage of the new method is that it can go through a large number of search results, fetching the page content, and seeing if it actually applies... much faster than I can. I get a much better starting point in much less time.

[jqpabc123]

I get a much better starting point in much less time.

So you're not actually reviewing the results? If you were, the time wouldn't be "much less".

You're getting something that you *assume* is a better starting point but you don't really know if the results are flawed or not.

If Google AI was telling people your mom was an ax murderer, would you *assume* it is providing a good starting point?

This is what AI does --- it lulls people into accepting bad results as a time saver. And this is the root of legal liability issues.

[verdverm]

I review the relevant results, instead of every search link that comes back, many of which are irrelevant. I also spend less time clicking in the browser.

I'm very much human-in-the-loop oriented. If you had seen any of my talks, you would understand this. Please refrain from making assumptions about how everyone uses Ai and putting them into a single classification. It's not what's going on.