Re:The People Who Ruined the Internet
by pclmulqdq on Thursday November 02, 2023 @12:31AM
(#38109015)
If you search "speed work running drills" on Google, you get relevant results. It kind of sounds like you are cherry picking the "bad prompts" you sent to Google and comparing them to good prompts sent to GPT-4.
[ Parent
| on HN ]
Re:The People Who Ruined the Internet
by lolinder on Thursday November 02, 2023 @01:50AM
(#38109447)
They specifically said that they used the same prompt. If I feed the exact string "speed work running" into ChatGPT 3.5 I get back essentially the same results that OP includes.
Re:The People Who Ruined the Internet
by pclmulqdq on Thursday November 02, 2023 @10:01AM
(#38113622)
Bad prompts for google can have exactly the same words as good prompts for ChatGPT. The original commenter probably uses both of them at least 10 times a day, so it's not unreasonable to assume that they will run into those examples.
Re:The People Who Ruined the Internet
by lolinder on Thursday November 02, 2023 @11:00AM
(#38114425)
I would typically assume that a three word prompt like that is better for Google than it is for ChatGPT. Google is supposed to be good at keyword search, and ChatGPT is designed for longer prompts.
Re:The People Who Ruined the Internet
by pclmulqdq on Thursday November 02, 2023 @11:53AM
(#38115304)
That means that your prior assumptions about Google are wrong. Google does a substantial amount of "semantic understanding" before doing a search. It is no longer a keyword matching tool - that stopped being the case a long time ago.
Also, ChatGPT is usually really bad with long prompts when I use it, and is often kind of close when I give it a short prompt. I assume part of this effect is psychological though, where you expect the short prompt to have worse results.