The phenomenon of correcting people because Chat Gippity said x or y was the beginning. Now people repeat what the machine said as though it originated from them, and this has been totally normalized. It permeates everything. People feel empowered by it, but they have no intent or ability to verify. This is normal. It's another source of information, but it's vetted by probability at best, yet also misinterpreted and internalized at worst.

People plagiarize and behave as though it's their own work with total confidence and no shame whatsoever. Speaking to teachers about this is mind-blowing. This is very real and present. These people believe they're doing 'the work' in many cases. Some are aware it's a farce, many are not.

It has jammed a lever into D-K and cranked it up into something even worse, in my opinion.

The giveaway was my Medical Professional father thinking that AI was really good at things outside of his area of expertise, and really bad at things inside of his area of expertise.

There’s a subtlety here, tho. It doesn’t mean (at all) that’s it useless. Just that those who don’t know what they’re doing are going to be amplifying their incompetence.

If it only amplifies half the effect, I don’t think TFA is an accurate claim.

https://news.ycombinator.com/item?id=45876744 - "LLMs are steroids for your Dunning-Kruger" (bytesauna.com)

392 points | 7 months ago | 301 comments

https://news.ycombinator.com/item?id=45851483 - "AI is Dunning-Kruger as a service" (christianheilmann.com)

268 points | 7 months ago | 199 comments

I recommend the book "The Information" (2011, James Gleick) for a grounded history of how we've slowly gone from using smoke signals and drums to transmit information to how we crossed the inflection point between being limited in how much information we could consume to instead being limited in how well we could curate the information we consume. Curation will continue to improve, and misplaced confidence is merely one downstream effect.

I also recommend, for those struggling with these effects, the book "Amusing Ourselves to Death" (1985, Neil Postman, updated in 2005 by his son), to understand the concept of Information Action Ratio - how to best target your curation.

Definitely is.

It has graduated into being an active and fairly precisely targeted thought shaping tool.

I worked on Ads and Feed Ranking at FB/IG and we never dreamed of the scope for shaping behavior and opinion that is now routinely deployed by frontier and near frontier vendors. RLHF is basically feed ranking in the first place, preference gradient with no ground truth referee, late SFT on amplifying data sets, and affine injections into the residual stream with a fluent, earnest base model that the public has been conditioned to regard as omniscient and wise?

Yeah that's fucking mind control when applied at scale. We did some sketchy shit a decade ago, this is next level.

Yeah we know that LLMs tend towards sycophancy.

Discussing DK has a real Matthew 7:3-5 vibe about it.