Here’s Siri's utterly botched result for “What was the Vikings-Bears score last week?”, and ChatGPT's response to the exact same query.
Siri is technically correct that the Vikings and Bears did not play between Dec 8 and last Saturday, Dec. 15, but they did play last week on Sunday. [Edit: Monday] ChatGPT completely nailed it, with the correct score in the first paragraph and more details below, including sources at ESPN and Reuters.
Asking a LLM for accurate answers is misusing it. It simply isn't in its nature to do that.
If you want accurate answers, you need an old fashioned hand coded application. It was designed for it.
LLMs can be good for suggestions, inspiration or first drafts.
By the way, there have been non-LLM applications for that since forever. But they have never had all the money LLMs have behind, so they have always been small tools.
A tailored app will always be better than a LLM for the same feature. The time to use an LLM is when it is the only viable option.
Then it is the search engine who is doing the work, so you may as well use directly the search engine and avoid any mistake the LLM may make. You will get the answers faster and with less power consumption, to boot.
And introducing hallucinations while doing it, forcing me to fact check everything.
I find much more useful, reliable and faster to use traditional search engines. At least, I know who said it, where, and have context. This is key if you need reliable information.
Except Google and Bing, of course, that have become total shit, very likely to make AI search look better.
Not that the Google Search of ten years ago weren't already using AI. Not an LLM, but AI nonetheless. And it was really great.
@jgg @gruber ChatGPT lists all sources for the answers. Did you actually use ChatGPT Search (https://openai.com/index/introducing-chatgpt-search/)? Sounds to me like you haven't heard about it until know.
@jgg @gruber The point is that the LLM acts as an agent doing the search for you, looking at the results and delivering answers to your questions in natural language.