Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

The rise and fall of Microsoft’s new Bing

garymarcus.substack.com

Post reply on HN
Print view View original

51–60 of 68 posts

Re: The rise and fall of Microsoft’s new Bing

#51

Post by fauigerzigerk »

mrtksn wrote: post #49
fauigerzigerk wrote: post #47

Earlier quoted context omitted.

>Factual correctness is not very important, you don't have to trust it. If these chat bots are going to replace or compete with the traditional search interface, then correctness is absolutely key. When a traditional search engine points you to some spammy site, you have far more context to decide for yourself how much credibility you ascribe to it. You may know the site from earlier visits. You can see whether it's…

I don't think its going to be drop-in replacement. With ChatGPT, you can ask follow up questions. So if something doesn't sound right, you can pressure it to explain its position. It's like talking with people, sometimes people are wrong and we happen to have a lot of linguistic tools to deal with it. Like, follow up questions or asking for specifics.

>I don't think its going to be drop-in replacement.

I agree, but part of the reason for that is that it doesn't get the facts right (yet?). A chat bot that can be trusted would be far more useful and could largely replace the traditional search interface.

I tried ChatGPT (and some other transformer models). I know you can ask follow up questions and make it reconsider its responses or change its mind.

But why does it change its mind? It's not because it finds a flaw in its own logic or discovers new information. I think it's simply because the pattern of inquisitorial questioning is statistically associated with the pattern of mind changing.

And then it just follows your lead in a new direction making claims that are just as unreliable as the original ones.

I'm keeping an open mind though. I don't think I have understood everything there is to understand about this.

Re: The rise and fall of Microsoft’s new Bing

#52

Post by NobleLie »

M4v3R wrote: post #30
Thorentis wrote: post #24

Earlier quoted context omitted.

> ChatGPT is very effective Massive [Citation Needed] to be honest. ChatGPT is arguably even worse than Bing at producing factually correct information, since it does not have any ability to source it. It simply predicts the next most likely word in a sentence. Even if that next word makes the entire thing horribly incorrect.

> It simply predicts the next most likely word in a sentence So does your brain. And yes, you are correct that at this moment ChatGPT is not very good at producing factually correct information on many topics, but that's not the only reason chatbots exist. It can do many more things, and I personally (as a programmer) and some people I know from different fields use it daily because it provides value for us (enough v…

Where did you learn that the brain works by simply predicting the next most likely word in a sentence?

Tertiary note: I'm also a programmer and a researcher. It's failed pretty hard on both. Faking sources of medical / biological information, botching simple chronology, suggesting made up methods on well known programing libraries (3 of many examples). There was nothing I could do to correct any of these errors - no matter how much leading or pointing out error, it would acknowledge it and repeat it.

I recognize the value of chatgpt (and roughly how it operates) but the excitement (marketing hype) is indeed over the top. It appears an incremental advancement of old word2vec (added human supervision and a few more things) but the sum total is still of questionable value in domains that mean the most to me (admittedly self centric, but many are like me, at least here)

Re: The rise and fall of Microsoft’s new Bing

#53

Post by college_physics »

herculity275 wrote: post #48
college_physics wrote: post #26

Language models feel currently a bit like blockchain: a solution looking for a problem. But unlike the latter where the actual problems are fairly niche and depend on myriad other constraints, its more likely that language algorithms will be adopted more widely. The question is how. People are currently slotting this tool into "search" (as we have come to know it) as it is obviously information retrieval machinery at…

I was recently reminded how in the Harry Potter books random appliances/tools/visual media all have a tiny bit of intelligence and can speak back to you. They typically have no agency and don't appear sentient, but they make the magical world a lot more interactive. I do wonder if the near-short term for LLM adoption looks a bit like that.

the experience with voice assistants seems to suggest that lowering expectations would be a good thing, as people are both gullible and ready to believe in magic and quickly come to get disappointed by fake magicians.

alas lowering expectations is exactly the opposite of VC culture.

Re: The rise and fall of Microsoft’s new Bing

#54

Post by hrunt »

mrtksn wrote: post #28
Thorentis wrote: post #24

Earlier quoted context omitted.

> ChatGPT is very effective Massive [Citation Needed] to be honest. ChatGPT is arguably even worse than Bing at producing factually correct information, since it does not have any ability to source it. It simply predicts the next most likely word in a sentence. Even if that next word makes the entire thing horribly incorrect.

Factual correctness is not very important, you don't have to trust it. ChatGPT is very effective because it acts as an advisor in any topic. Sometimes its very wrong but it gives you an overall picture of the thing you are looking for and from there you can drill down. It is much better than trying to drill down from spam ridden ad infested results Google would provide. Once you pass the very basic stuff, you are oft…

The scary thing about this reply is I cannot tell whether it was written by a human or by Bing trying to gaslight me into believing that factual correctness is not important. It reads just like Bing trying to explain why the year is 2022 or Bing did not say that it said the Eagles won the Super Bowl.

Re: The rise and fall of Microsoft’s new Bing

#55

Post by mrtksn »

hrunt wrote: post #54
mrtksn wrote: post #28

Earlier quoted context omitted.

Factual correctness is not very important, you don't have to trust it. ChatGPT is very effective because it acts as an advisor in any topic. Sometimes its very wrong but it gives you an overall picture of the thing you are looking for and from there you can drill down. It is much better than trying to drill down from spam ridden ad infested results Google would provide. Once you pass the very basic stuff, you are oft…

The scary thing about this reply is I cannot tell whether it was written by a human or by Bing trying to gaslight me into believing that factual correctness is not important. It reads just like Bing trying to explain why the year is 2022 or Bing did not say that it said the Eagles won the Super Bowl.

The scary part is that people can start dismissing the opinions that they don't agree as "AI generated".

Re: The rise and fall of Microsoft’s new Bing

#56

Post by bell-cot »

mrtksn wrote: post #55
hrunt wrote: post #54

Earlier quoted context omitted.

The scary thing about this reply is I cannot tell whether it was written by a human or by Bing trying to gaslight me into believing that factual correctness is not important. It reads just like Bing trying to explain why the year is 2022 or Bing did not say that it said the Eagles won the Super Bowl.

The scary part is that people can start dismissing the opinions that they don't agree as "AI generated".

People have been dismissing opinions they didn't like for ~25,000 years, for a very long list of reasons. Dismissing "AI" for that reason feels kinda like not having one more key on your keyring - because a kid could stick the new key into an electrical outlet.

Re: The rise and fall of Microsoft’s new Bing

#57

Post by mrtksn »

bell-cot wrote: post #56
mrtksn wrote: post #55

Earlier quoted context omitted.

The scary part is that people can start dismissing the opinions that they don't agree as "AI generated".

People have been dismissing opinions they didn't like for ~25,000 years, for a very long list of reasons. Dismissing "AI" for that reason feels kinda like not having one more key on your keyring - because a kid could stick the new key into an electrical outlet.

Dismissing opinions is on thing, dismissing opinions thinking that you are not interacting with humans is why some people are huge a-holes on the internet despite being decent people IRL.

Re: The rise and fall of Microsoft’s new Bing

#58

Post by bell-cot »

mrtksn wrote: post #57
bell-cot wrote: post #56

Earlier quoted context omitted.

People have been dismissing opinions they didn't like for ~25,000 years, for a very long list of reasons. Dismissing "AI" for that reason feels kinda like not having one more key on your keyring - because a kid could stick the new key into an electrical outlet.

Dismissing opinions is on thing, dismissing opinions thinking that you are not interacting with humans is why some people are huge a-holes on the internet despite being decent people IRL.

Maybe I'm an old cynic...but I already (quietly) dismiss most plausibly-human opinions on the internet, because "advertising", "clickbait", "ignorant", "echo chamber", "troll", "spam", "obvious & boring", "not my religion", "immature". Etc. etc. etc.

"Is it from an AI?" seems about as important to me as validating the SPF record behind some "I am assistant to Nigerian Minister of Oil, and need your help..." email message. Who would actually react differently to that message, on the basis of "SPF is valid"?

Re: The rise and fall of Microsoft’s new Bing

#59

Post by mrtksn »

bell-cot wrote: post #58
mrtksn wrote: post #57

Earlier quoted context omitted.

Dismissing opinions is on thing, dismissing opinions thinking that you are not interacting with humans is why some people are huge a-holes on the internet despite being decent people IRL.

Maybe I'm an old cynic...but I already (quietly) dismiss most plausibly-human opinions on the internet, because "advertising", "clickbait", "ignorant", "echo chamber", "troll", "spam", "obvious & boring", "not my religion", "immature". Etc. etc. etc. "Is it from an AI?" seems about as important to me as validating the SPF record behind some "I am assistant to Nigerian Minister of Oil, and need your help..." email mes…

Yep, in old generations there was this thing about internet being a new frontier where you can nerd down and it wasn't a place for "normal" people and whatever happens online doesn't have IRL consequences. Some of them turned into unbearable a-holes, however newer generations have another kind of relationship with the internet where they are not assuming that everyone is a neckbeard and there are no women on the internet etc. Instead, they are much less anonymous, use real names and are not in the delusion that there are no real world consequences. It's the only way the normies to stick around, despite having orders of magnitude more people online the trolling is not as pervasive as it used to be.

Re: The rise and fall of Microsoft’s new Bing

#60

Post by Slava_Propanei »

mrtksn wrote: post #55
hrunt wrote: post #54

Earlier quoted context omitted.

The scary thing about this reply is I cannot tell whether it was written by a human or by Bing trying to gaslight me into believing that factual correctness is not important. It reads just like Bing trying to explain why the year is 2022 or Bing did not say that it said the Eagles won the Super Bowl.

The scary part is that people can start dismissing the opinions that they don't agree as "AI generated".

[dead]
Post reply on HN