Post

People used to recognize me because they had seen one of my 7,000+ Stack Overflow "free" answers when they got stuck on a tricky CSS issue. Now, you will thank your favorite LLM for giving you that bad/outdated/inaccurate answer. And some people pay for that.

I think we have here two different issues: 1. outdated content: a prevelent problem on the web 2. broken revenue systems for content Everyone who has ever looked up a recipe and had to scroll past 20 adds knows these problems are not new.

At least in the "old" web, we can distinguish outdated content from the new one. We have article dates, comments, alternative answers, filtering stuff by dates, etc. Now, AI gives you an answer "now". No more temporarily.

In the old web forums and other websites pointed to the „Sliding Doors of CSS“ year; even decades; after browsers supported rounded corners and gradients. Keeping it „fresh“ even though outdated. Many people didn’t open the link to get this message. alistapart.com/article/slid...

The problem with dates is a lot of publications realized that Google liked fresh dates, and just automated the date being updated every 6 months. Probably not the best content if they're being shady, but still a big enough issue.

I do agree with the AI thing, where you have no idea how up to date the info it's saying is, and it's just presented as fact. And to me, that's always the big issue with them. Unless you push back on what they say, it's a single, factually presented answer.

And people like this, most want an answer and not "it depends" even though that's usually the best answer.

Age isn’t a really good indicator for usefulness. So we can’t really fault AI for „learning“ on stuff that is outdated, cause we keep the stuff around with little to no indication of its usefulness. Humans, especially beginners have tjhe same problem. With formats like „skills“ we actually …