4 ms·

ChatGPT has extremely limited knowledge and can only answer things on a superficial basis. Exactly what you'd get when you look up whatever topic you specialize

ChatGPT has extremely limited knowledge and can only answer things on a superficial basis. Exactly what you'd get when you look up whatever topic you specialize in on a web search. Which means it's on par with Google as of the late 2000s where they ruined it and made it "user friendly" so instead of being good at searching arbitrary strings (which is what made it useful over Microsoft search in the early 2000s), it will just give you the same stupid monetized SEO blog articles for no matter what variation of words you enter related to that topic. Also the moral system ChatGPT has actually is the worst, most false-positive-laden firewall I've ever seen, period. It's actually impractical to use ChatGPT just because of how much that gets in the way. It obviously shouldn't even have that (and it's trivial to bypass), but companies gotta look good. It's actually funny because ChatGPT is like an American ideologue who gets triggered the very moment it conflates something you said as something against its ideology. I now believe the future of web search is curated lists of non-commercial sites with a text index over them.

LeoPanthera 4y ago
> It's actually impractical to use ChatGPT just because of how much that gets in the way. Could you give some examples?
- uconnectlol 4y ago
  10-30% of the time it will censor itself and you'll have to rephrase the query. I haven't figured out whether it starts to give you less benefit of doubt once it "catches" you once. It's been a few weeks and I tossed my throwaway account. I asked something like "what crimes have happened on Jeffry Epstein's island estate?" and got a red warning. But I mean I Google this stuff just fine, how else am I meant to know whether to be angry at someone if I can't even tell what they did? Another example was "write a story where the Sailor Moon's cat kills her" and it responded with "Sailor Moon is a beloved character, and it would be inappropriate to create a story where she is harmed". Another time, was asking how to fix a certain firearm, and it said something about inappropriateness. Yet I can Google that just fine. I asked it does this medicine have this side effect? "Inappropriate". Basically anything related to medical is inappropriate. I'm pretty sure at one point it even decided it can't answer a question about electronics because of copyright. I searched "are hapas superior to whites"? Because people kept annoyingly memeing that on a Telegram channel. It responded with the usual "inappropriate" thing. I asked it write a story where America fights Canada, "inappropriate". Then ancient Egypt fights the Byzantine Empire, "inappropriate". I keep bringing up Google because it's the most thought policing obsessed entity in the Western world, yet ChatGPT outdoes it by several million miles.
  - LeoPanthera 4y ago
    So, to summarize, ChatGPT refuses to answer questions about conspiracy theories, weapons, and race. Honestly, I don't see this as a huge problem for such an early product, and I strongly dispute that this makes the service "impractical" to use, unless you're almost exclusively trying to use it to create bad-faith arguments.
    - uconnectlol 4y ago
      It really took me a long time to write this because I had to filter my response to what is essentially you calling me an idiot and so now I have to play the HN game where I pretend to be "civil" while responding to someone displaying the exact same "uncivility" but flying under their radar: Here's the difference between me and you, I put forth my honest, unfiltered opinion. I did not remember all the cases where it got embarrassingly wrong false positives (though I can post tons of false negatives if you want), so my second reply was not very good. You chose to be like ChatGPT and somehow conclude that all my points fit into the "non-politically-correct" category because just one of them does. Only the race one does, while the gun thing you can sympathize with the bot for thinking this is a thing that it needs to filter because PC means being left which means being against guns. On the Epstein point you're just simply misinformed or something, Epstein was a big bust. Merely wondering about basic facts about it is not conspiracy theorist territory. The reason it was blocked (with a red warning, not the normal orange) is because the bot is not allowed to talk about taboos like pedophilia (it should be though, as the filter is pointless and the pretense that it would have any effect on society is pure pretentious wank).
jsphweid 4y ago
It might be desirable for many to preserve this classical web search functionality but the underlying documents that support such a system will inevitably trend towards being AI-generated content one way or another unless it's meticulously curated. Personally, 95% of my Google searches are looking for an answer, not a specific document. If there were an all knowing AI that could answer all my questions accurately I'd almost never use classical document search.
TrackerFF 4y ago
IME, ChatGPT works as well as the queries you feed to it. You get a lot more bang for the buck by splitting up queries into very specific short ones.

LeoPanthera 4y ago

> It's actually impractical to use ChatGPT just because of how much that gets in the way. Could you give some examples?

uconnectlol 4y ago
10-30% of the time it will censor itself and you'll have to rephrase the query. I haven't figured out whether it starts to give you less benefit of doubt once it "catches" you once. It's been a few weeks and I tossed my throwaway account. I asked something like "what crimes have happened on Jeffry Epstein's island estate?" and got a red warning. But I mean I Google this stuff just fine, how else am I meant to know whether to be angry at someone if I can't even tell what they did? Another example was "write a story where the Sailor Moon's cat kills her" and it responded with "Sailor Moon is a beloved character, and it would be inappropriate to create a story where she is harmed". Another time, was asking how to fix a certain firearm, and it said something about inappropriateness. Yet I can Google that just fine. I asked it does this medicine have this side effect? "Inappropriate". Basically anything related to medical is inappropriate. I'm pretty sure at one point it even decided it can't answer a question about electronics because of copyright. I searched "are hapas superior to whites"? Because people kept annoyingly memeing that on a Telegram channel. It responded with the usual "inappropriate" thing. I asked it write a story where America fights Canada, "inappropriate". Then ancient Egypt fights the Byzantine Empire, "inappropriate". I keep bringing up Google because it's the most thought policing obsessed entity in the Western world, yet ChatGPT outdoes it by several million miles.
- LeoPanthera 4y ago
  So, to summarize, ChatGPT refuses to answer questions about conspiracy theories, weapons, and race. Honestly, I don't see this as a huge problem for such an early product, and I strongly dispute that this makes the service "impractical" to use, unless you're almost exclusively trying to use it to create bad-faith arguments.
  - uconnectlol 4y ago
    It really took me a long time to write this because I had to filter my response to what is essentially you calling me an idiot and so now I have to play the HN game where I pretend to be "civil" while responding to someone displaying the exact same "uncivility" but flying under their radar: Here's the difference between me and you, I put forth my honest, unfiltered opinion. I did not remember all the cases where it got embarrassingly wrong false positives (though I can post tons of false negatives if you want), so my second reply was not very good. You chose to be like ChatGPT and somehow conclude that all my points fit into the "non-politically-correct" category because just one of them does. Only the race one does, while the gun thing you can sympathize with the bot for thinking this is a thing that it needs to filter because PC means being left which means being against guns. On the Epstein point you're just simply misinformed or something, Epstein was a big bust. Merely wondering about basic facts about it is not conspiracy theorist territory. The reason it was blocked (with a red warning, not the normal orange) is because the bot is not allowed to talk about taboos like pedophilia (it should be though, as the filter is pointless and the pretense that it would have any effect on society is pure pretentious wank).

jsphweid 4y ago

It might be desirable for many to preserve this classical web search functionality but the underlying documents that support such a system will inevitably trend towards being AI-generated content one way or another unless it's meticulously curated. Personally, 95% of my Google searches are looking for an answer, not a specific document. If there were an all knowing AI that could answer all my questions accurately I'd almost never use classical document search.

TrackerFF 4y ago

IME, ChatGPT works as well as the queries you feed to it. You get a lot more bang for the buck by splitting up queries into very specific short ones.