4 ms·

I believe Google is going to see a major hit and the thing is they can't do anything to stop it. They have an incentive to keep search traffic high and unfortu

by taf2 4y ago

I believe Google is going to see a major hit and the thing is they can't do anything to stop it. They have an incentive to keep search traffic high and unfortunately for them GPT-4 is so good that it gets answers faster with less ads then Google. It's similar to how Google disrupted other search engines in the late 90s. I'm thinking about how I craft my prompt to get the right answer, similar to how I used to think about how I craft the correct keyword search to get the right answer. I have not even tried via bing.com yet just using chat.openai.com selecting gpt-4 and I find much better answers faster then googling... I still google somethings like: "convert 70F to C"... but for help with coding solutions I just ask gpt4

nicce 4y ago
> GPT-4 is so good that it gets answers faster The problem is, that if no references are provided, then you need to manually verify the content… by using search engine.
- taf2 4y ago
  For code - I can read the code and verify myself... If I was asking whether the sky is blue or some other kind of "fact" based question sure... But really how is that different with search? I can look at the website and say ah I trust this website over that other website... I think it's a new flavor of "don't trust everything you hear on the TV" Or in more modern day "Don't trust everything you read on the internet", or "Don't trust everything you hear on social media"... It's just the next iteration... "Don't trust everything an AI tell you". That is no different from me learning from code i find via search but instead ask about via AI...
  - nicce 4y ago
    > "Don't trust everything you read on the internet", or "Don't trust everything you hear on social media" There is a difference. There are places with reputation and backlogs and even with scientific references. But if AI just gives you a text, there is nothing. Until they fix references, you can’t really use it for anything factual, non-logical information.
    - taude 4y ago
      If these were the case, then why would "fake news" be a thing?
      - nicce 4y ago
        Fake news are exactly the current state of the A.I. presentation. Confident expressions without proper referencing. People also like that their own opinions are validated, regardless of the truth. Which is also another thing why fake news success. I would say that fake news is an excellent example why current Chatbots might not be so good, as they express information in similar way.
- nullsense 4y ago
  All this will be a solved problem soon enough.
- JumpCrisscross 4y ago
  > you need to manually verify the content Why wouldn’t the chat provide you with sources on request? They seem to be doing this e.g. with the Wolfram integration.
  - nicce 4y ago
    > Why wouldn’t the chat provide you with sources on request? Because many are still fake, nonexistent. Wolfram is very limited scope for the global population needs. E.g if you ask about local hunting laws from ChatGPT, without correct reference, what can you do with that information? It is very dangerous to trust them blindly.
    - pwinnski 4y ago
      Bing is exposing source links for their answers, one-click verification. Wolfram has a much wider scope of information than I think most people realize, but your example of local hunting laws is a good example of what they don't cover. I asked Wolfram, and it told me how many calories I'd burn while hunting. I asked Google, and the first link was to Hunting Regulations - Outdoor Annual - TPWD - Texas.gov, excellent result. I asked OpenAI, and it cautioned that it isn't up-to-date, since it's just a language model, advised that TPWD is authoritative (correct), and gave an accurate summary of the five major point of gun laws in Texas before reminding me to check with TPWD for the latest. I've had some misses with OpenAI, but this was an excellent answer. I've also had some misses with Google, but this turned out to be something they surfaced excellent link for, too.
      - nicce 4y ago
        > I've had some misses with OpenAI, but this was an excellent answer. I've also had some misses with Google, but this turned out to be something they surfaced excellent link for, too. Think about the bias - maybe you were able to verify that the answer was excellent because you already knew the laws? Or you Googled and compared it? In both cases, ChatGPT did not perfom alone in providing validity. It kinda gave a summary for search results of Google, or for your pre-knowledge. Imagine if someone who does not know the laws, asks about them. ChatGPT cannot offer any validity alone for the user, at least yet. ChatGPT provides often excellent answers, I am not denying it. But the bad 5% ruins everything if they sound alike for the correct 95%, and if ChatGPT cannot provide any validity for them, at least yet. Often you ask things you don’t know or haven’t searched before.
- huseyinkeles 4y ago
  Bing chat is using GPT-4 and does provide sources. Also, it has access to the internet, so I am able to ask about very recent movies etc.
  - nicce 4y ago
    Didn’t it just cite Hacker News comment and Bard… there is still some work to do.
    - huseyinkeles 4y ago
      Well, you still need to do your diligence. At least you have a link to the source.

by taf2 4y ago

nicce 4y ago

> GPT-4 is so good that it gets answers faster The problem is, that if no references are provided, then you need to manually verify the content… by using search engine.

taf2 4y ago
For code - I can read the code and verify myself... If I was asking whether the sky is blue or some other kind of "fact" based question sure... But really how is that different with search? I can look at the website and say ah I trust this website over that other website... I think it's a new flavor of "don't trust everything you hear on the TV" Or in more modern day "Don't trust everything you read on the internet", or "Don't trust everything you hear on social media"... It's just the next iteration... "Don't trust everything an AI tell you". That is no different from me learning from code i find via search but instead ask about via AI...
- nicce 4y ago
  > "Don't trust everything you read on the internet", or "Don't trust everything you hear on social media" There is a difference. There are places with reputation and backlogs and even with scientific references. But if AI just gives you a text, there is nothing. Until they fix references, you can’t really use it for anything factual, non-logical information.
  - taude 4y ago
    If these were the case, then why would "fake news" be a thing?
    - nicce 4y ago
      Fake news are exactly the current state of the A.I. presentation. Confident expressions without proper referencing. People also like that their own opinions are validated, regardless of the truth. Which is also another thing why fake news success. I would say that fake news is an excellent example why current Chatbots might not be so good, as they express information in similar way.
nullsense 4y ago
All this will be a solved problem soon enough.
JumpCrisscross 4y ago
> you need to manually verify the content Why wouldn’t the chat provide you with sources on request? They seem to be doing this e.g. with the Wolfram integration.
- nicce 4y ago
  > Why wouldn’t the chat provide you with sources on request? Because many are still fake, nonexistent. Wolfram is very limited scope for the global population needs. E.g if you ask about local hunting laws from ChatGPT, without correct reference, what can you do with that information? It is very dangerous to trust them blindly.
  - pwinnski 4y ago
    Bing is exposing source links for their answers, one-click verification. Wolfram has a much wider scope of information than I think most people realize, but your example of local hunting laws is a good example of what they don't cover. I asked Wolfram, and it told me how many calories I'd burn while hunting. I asked Google, and the first link was to Hunting Regulations - Outdoor Annual - TPWD - Texas.gov, excellent result. I asked OpenAI, and it cautioned that it isn't up-to-date, since it's just a language model, advised that TPWD is authoritative (correct), and gave an accurate summary of the five major point of gun laws in Texas before reminding me to check with TPWD for the latest. I've had some misses with OpenAI, but this was an excellent answer. I've also had some misses with Google, but this turned out to be something they surfaced excellent link for, too.
    - nicce 4y ago
      > I've had some misses with OpenAI, but this was an excellent answer. I've also had some misses with Google, but this turned out to be something they surfaced excellent link for, too. Think about the bias - maybe you were able to verify that the answer was excellent because you already knew the laws? Or you Googled and compared it? In both cases, ChatGPT did not perfom alone in providing validity. It kinda gave a summary for search results of Google, or for your pre-knowledge. Imagine if someone who does not know the laws, asks about them. ChatGPT cannot offer any validity alone for the user, at least yet. ChatGPT provides often excellent answers, I am not denying it. But the bad 5% ruins everything if they sound alike for the correct 95%, and if ChatGPT cannot provide any validity for them, at least yet. Often you ask things you don’t know or haven’t searched before.
huseyinkeles 4y ago
Bing chat is using GPT-4 and does provide sources. Also, it has access to the internet, so I am able to ask about very recent movies etc.
- nicce 4y ago
  Didn’t it just cite Hacker News comment and Bard… there is still some work to do.
  - huseyinkeles 4y ago
    Well, you still need to do your diligence. At least you have a link to the source.