I started to write this blog post a few months ago, and just published a post pontificating / complaining about AI search result slop.
Google has become nearly useless lately. Searching for anything turns up it’s AI slop nonsense or advertisements rather than actual links to something useful. Google’s initial promise – high quality uncluttered search results is dead. Any results are now hidden below their AI slop, advertisements, and other results from other slop sources. I might as well publish it now, rather than have it relegated to a draft for years to come.
If it were possible for Google AI summaries of search results to be infallible, I guess that would be fine – but it’s just so far from the truth. I’ve yet to see an LLM that doesn’t just try to generate something to please – rather than try to generate something that is accurate. Google AI offering false but pleasing slop reminds me of the Azimov story Liar!. Without a “three laws of robotics” style filter, there’s no systemic guardrail preventing automated misinformation.
While I think there’s room for legitimate practical LLM uses, shoving it into search results seems like a complete waste for everyone concerned. When people say “ChatGPT/Claude told me X”, that’s really no different than saying “ChatGPT/Claude thought I would be pleased if it said X.” But using an LLM1 for synthesis and summarization of information and data seems only able to generate either results its hopes are pleasing to the requestor or results hopelessly biased by it’s training. When too many people have the wrong idea about what an LLM is, how it works, or how to interpret it’s responses, putting an AI result before actually informative links is really damaging.
Grumble, grumble, get off my lawn.
- At least, as they exist now [↩]