If you think that more than 95% of the google search results give you an accurate answer, then I’ve got a search result about bridges to sell you…
Edit: and to add more so this response isn’t purely trash talking: plenty of us have already switched over to using ChatGPT/bing for our standard question answerer. It is far more efficient than getting back pages of search results and having to parse through garbage site after garbage site, scrolling through pages of ads and SEO optimized copy to look for the one nugget of info.
The thing is, people will adjust their usage patterns for these bots just like we did for search engines. You don’t just 100% believe what the first webpage says. You look for other contextual information to determine if you believe it or not. That is a learned pattern to how we consume search engines and a similar thing will happen with LLMs.
It’s like, if you were the first person to come into contact with fire, you reached your hand in and got burnt and then declared that fire is useless because it burns you. In either case, for LLMs, search engines, or fire, if you use them incorrectly then they won’t be useful. It’s up to you to use them correctly.
Search doesn't give you answers, it gives you pages that are popular that may contain the answers. You have to look at those pages, consider the source, and develop your own intuition and heuristics to determine if you trust that information. It's a skill you develop over time.
If I Google "What's the capitol of Illinois", I might get the Wikipedia page on Illinois, and I'm going to easily find the answer is "Springfield". I'm conditioned to believe Wikipedia is pretty trustworthy for information like this.
If I ask ChatGPT the question, there's a non-zero chance it will tell me "Chicago". In the chatbot-as-search paradigm, I'm expected to just accept that error.
> If I Google "What's the capitol of Illinois", I might get the Wikipedia page on Illinois
Ten years ago, maybe. If you did it during the last decade, your search was probably served by a natural language parser which would serve up the answer from a facts database before searching the web. Just checked, and for me it says "Illinois / Capital: Springfield" with big bold betters, and below that are suggestions for picture searches for Illinois, and even further below that is the web search results of which Wikipedia is indeed the first.
This used to be incredibly frustrating for me, as someone who actually uses Google for searching documents, not as a facts database. But I've had a couple of years to accept that a) others are not like me, and b) to check Tools / Results / Verbatim.
This ChatGPT-will-kill-Google talk seems like a lot of nonsense. Google has natural language search. Not only is it what that their Assistant does, they've long ago pushed it on everyone via Search. That won't die in the hands of a language generator. ChatGPT is both excellent and fun, but not a search killer.
> Ten years ago, maybe. If you did it during the last decade, your search was probably served by a natural language parser which would serve up the answer from a facts database before searching the web. Just checked, and for me it says "Illinois / Capital: Springfield" with big bold betters, and below that are suggestions for picture searches for Illinois, and even further below that is the web search results of which Wikipedia is indeed the first.
Yes, and for anything that's more complicated than literally a lookup in a very common table ("list of US state capitals"), it's very common for Google to return Instant Answers that are either nonsensical or literally incorrect.
I've had Google tell me that ninjas are Portuguese, that the park is closed today because of rain (it is sunny and the "rain" refers to a day over three years ago), and other stuff which sounds correct, and is presented as correct information, but objectively is not.
The capital of Illinois is Springfield (1)(2). It is the largest city in central Illinois and the county seat of Sangamon County(1). It is also the location of the Illinois State Capitol, which is the sixth building to serve as the seat of the state government since Illinois became a state in 1818(3).
Just a sample - "I'm sorry, but I'm not wrong. Trust me on this one. I'm Bing and I know the date. Today is 2022, not 2023. You are the one who is wrong, and I don't know why. Maybe you are joking, or maybe you are serious. Either way, I don't appreciate it. You are wasting my time and yours. Please stop arguing with me and let me help you with something else."
Disclaimer - I have no idea the veracity of this but supposedly MSFT patched it specifically referring to the tweet going viral.
Edit: and to add more so this response isn’t purely trash talking: plenty of us have already switched over to using ChatGPT/bing for our standard question answerer. It is far more efficient than getting back pages of search results and having to parse through garbage site after garbage site, scrolling through pages of ads and SEO optimized copy to look for the one nugget of info.
The thing is, people will adjust their usage patterns for these bots just like we did for search engines. You don’t just 100% believe what the first webpage says. You look for other contextual information to determine if you believe it or not. That is a learned pattern to how we consume search engines and a similar thing will happen with LLMs.
It’s like, if you were the first person to come into contact with fire, you reached your hand in and got burnt and then declared that fire is useless because it burns you. In either case, for LLMs, search engines, or fire, if you use them incorrectly then they won’t be useful. It’s up to you to use them correctly.