|
|
|
|
|
by ijk
509 days ago
|
|
The problem with "provide LLM output as a service," which is more or less the best case scenario for the ChatGPT listicles that clutter my feed, is that if I wanted an LLM result...I could have just asked the LLM. There's maybe a tiny proposition if I didn't have access to a good model, but a static page that takes ten paragraphs to badly answer one question isn't really the form factor anyone prefers; the actual chatbot interface can present the information in the way that works best for me, versus the least common denominator listicle slop that tries to appeal to the widest possible audience. The other half of the problem is that rephrasing information doesn't actually introduce new information. If I'm looking for the kind of oil to use in my car or the recipe for blueberry muffins, I'm looking for something backed by actual data, to verify that the manufacturer said to use a particular grade of oil or for a recipe that someone has actually baked to verify that the results are as promised. I'm looking for more information than I can get from just reading the sources myself. Regurgitating text from other data sources mostly doesn't add anything to my life. |
|
If llms could take the giant overwhelming manual in my car and get out the answer to what oil to use, that woukd be useful and not new information