Gregory's Server@rodhilton
If you ask it ChatGPT the answer is perfect
@zandbelt
|
15 comments
@ErikJonker @rodhilton I still use Google though...(and sometimes DuckDuckGo but should perhaps be the other way around) I am open for suggestion for better search engines ! Not fully convinced re: Qwant (don't know why) @zandbelt @rodhilton ... I like perplexity.ai , in addition to Google, yes it's an LLM and can't be trusted, you have to check results, but it provides sources to make that easier  @ErikJonker @zandbelt @rodhilton If you have enough knowledge to check the result, fine. But (a) people are lazy (proofs already exist of people who know but didn't check); and (b) what about the people who don't know any better? And when these systems start eating their own garbage it will just escalate into utter insanity. @adaddinsane @zandbelt @rodhilton ...the assumption is I think (which might be correct) that these tools will get better, ofcourse mistakes will remain but that's also possible with traditional search @ErikJonker @zandbelt @rodhilton @whole1day @zandbelt @rodhilton well perplexity annotates results with footnotes which are very often scientific papers @ErikJonker @rodhilton more or less the same that applies to ChatGPT, luckily you are still required to think and judge for yourself on top what is generated and presented faster then you can manage (but it is also a bit scary when you start thinking whether others will judge the results with care or just take things for granted...am I to cynical here? or in other person's view the guy that feels he is the genius and elite while others are just "stupid" idiots? Hope not...it IS quite scary?!) @ErikJonker @rodhilton @zandbelt Try asking it what weighs more: a ton of iron, or two tons of feathers. @adaddinsane @rodhilton @zandbelt ..that is not a question you should ask an LLM like ChatGPT , i would not use it for any math or arithmetic, with the exception of explaining theory for example (which is wel described in texts that are part of the trainingsset). Wolfram Alpha is good choice for algorithms in that context which is also a plugin of GPT4 (paid version) @ErikJonker @rodhilton @zandbelt Oh right, so where is the definitive list of what questions are allowed? (Sarcasm) The point is it gets it wrong because the "two" doesn't match the pattern it expects, so it just throws away the most critical part of the question. They're crap. They won't get better. The more restrictions they put in place manually the worse the side effects. Ask any programmer. @ErikJonker @rodhilton @zandbelt so if you ask it the prompt that gave that response that Googles quoting, it still gets it wrong
|
@ErikJonker @rodhilton that might be but the Google algorithm is definitely broken (that is the algorithm it serves for me), given the first two nonsense hits that show up. Google once used to be of higher quality search results....a pity.