I agree with this, but there's still a problem with how LLMs too often provide users information that contains a liberal or "goodthinky" framing.
Let me give you the most recent example I ran into: Last night, I asked Gemini for the mean SAT score across all HBCUs. (I knew what the score was, but I wanted confirmation before posting a tweet.)
Instead of giving me what I asked for, it provided a list of the five HBCUs with the highest average SATs, along with brief descriptions of the "excellence" of each of these schools.
So, I did what I always do, which is call it out. I asked: "Why did you provide the average student SAT scores for the top 5 HBCUs, and not the mean of all of them? Why did you do this?"
Gemini responded the way it always does, with an effusive apology, admitting its mistake, and then it provided a figure.
But the figure was the top of the typically stated mean range, not the actual mean. So, again, I had to call it out, and, again, it apologized and then gave me the actual mean.
I guess what I'm saying is, yes, LLMs are likely to have a positive effect countering "fake news" and conspiracy theories, but it would be nice if they didn't so frequently resort to unhelpful "goodthink" on some of the more contentious and "sensitive" culture war issues (like those relating to race and gender).
Finally, I'll add this: They all do seem to be getting less biased and more nuanced as time goes on. Three years ago, it was pointless asking most of them questions on certain topics, but these days, although sometimes only after some prodding, you're likely to get what you need within a reasonably fair and balanced presentation.
Steven Pinker @sapinkerAI could push back against public irrationality like conspiracy theories, fake news, & medical quackery. Whereas social media optimize for sheer engagement (clicks, likes, & reposts), AI is sold as a useful product that aids decision-making in the real world. For all the problems of LLMs, they're less likely to send users down a conspiratorial rabbit hole. This may restore the influence of epistemically responsible expertise. So argues the philosopher Dan Williams @danwilliamsphil
105652.1K







