Should AI generated posts be banned or otherwise regulated?

@hilde45 - I hear ya, but all the doom and gloom, sky-is-falling, fear-mongering rhetoric out there lately seems to be unnecessarily oppressive.  I get that many don't like change, but I refuse to live by worrying about what might happen, especially when much of it never does. I certainly understand why folks from Missouri say "Show Me". 

@mitch2 

I just read an article showing the people that are using AI the most are the higher level employees. They are typically overwhelmed with work, motivated and open to new ways of becoming more effective. So they worked their way up by hard work and are now becoming more productive... I’m guessing they will stay.

I’ve spent most of my career introducing new technology. There are good reasons to expect some differences... but it will also enable folks that adopt and use new tools to become incredibly productive. I have helped and seen many clerks and stock people go from $35K jobs doing mundane work to making well over $100K by being early adopters and then teaching and implementing new tech. Change affords great opportunities for those who grab on. '

Now alignment... that could be a showstopper. Fingers crossed on that one.  

I have helped and seen many clerks and stock people go from $35K jobs doing mundane work to making well over $100K by being early adopters and then teaching and implementing new tech.

Employee with paycheck makes $100K a year! You don't say!! Now that's an American success story if I ever seen one. Move over, Elon.

If you folks’ could answer the question, will the S&P close lower or higher tomorrow than today, you’d be freakin billionaires. But you can’t, so you’re not. Yet y’all pontificate about what AI will do to society a quarter century from now. Good luck.

Nothing is ever quite as it seems. 

@hilde45 One thing I’ll keep turning over: namely that LLM users weren’t just equivalent to the control group on condition identification -- they were measurably worse. Unless I’m confused, that finding alone seems like the sharpest challenge to the "skilled users will figure it out" interpretation, and I’m not sure the thread has fully reckoned with it. But it’s time for a walk.

We should look at the big picture—i.e., 95% accuracy by LLMs alone versus 34% when used by end users for condition identification. This suggests that AI is highly sensitive to how people ask questions and to the quality of those questions.

There’s nothing wrong with trying to rationalize why Google/web search by end users (42%) slightly outperforms LLM use by end users (34%). One possible reason could be, when performing web searches, people tend to refine their queries and compare or cross-reference multiple sources, which may improve accuracy relative to relying on a single synthesized answer from an LLM.

That said, we’re looking at results that, although statistically significant, are only marginally better (42% vs. 34%). Compared to the 95% accuracy achieved by LLMs alone, however, this represents a substantial (and disappointing) drop-off. It’s important to keep the big picture in perspective.