• I didn’t say they don’t serve a purpose, I said that customers didn’t ask for it/don’t want it. I also said AI, not LLM (initially).

      They serve a purpose, but in many cases are not well suited or ready for production.

      Examples?

      I use a food delivery app that added an AI chatbot to supposedly help you with restaurant and food suggestions, and it wasn’t able to handle simple dietary restrictions (whereas standard filtering options on the search work just fine). A prompt as simple as “I’m looking for restaurants with vegan sandwiches” would return places that had vegan food and/or sandwiches. Even if it worked properly, just having adequate filter options and robust search is far more suited to the kinds of things users are doing on this kind of app.

      Search engines putting their in-house LLM generated slop above actual search results, which is the equivalent of ensuring that the first search result is from a garbage site with hit or miss accuracy of information or relevance to the topic. I don’t imagine it will be long before we start to see generative ai image results being served on the fly by in-house products when you search for images. Pass the user query to an LLM to generate a prompt for image gen, then pass that to image gen and bam! Slop!

      Every fucking product has some stupid gradient-coloured ✨ AI button shoved into it now and it drives me up the wall.

      Putting LLM auto complete into IDEs? Great, as long as I can turn it off and as long as they’re not training it on the code I write in the IDE. AI denoising in rendering software? Excellent. Image infill/outfill using AI in photo editing and drawing software? Spot on.