4 ms·

I’ve seen a lot of articles criticizing these models for getting information wrong. I understand there is a lot of hype, and that choosing the hate train or the

I’ve seen a lot of articles criticizing these models for getting information wrong. I understand there is a lot of hype, and that choosing the hate train or the hype train will gets you lots of press, but does anyone actually believe these models will be right most of the time? It’s just another tool that can be used when researching or trying to understand something. You still have to do your due diligence and evaluate the information. That has always been the case, even with traditional search.

vsareto 4y ago
Once they became a serious product pushed by major tech companies, people aren't unreasonable in expecting it to be right most of the time. Especially if it's going to go alongside or replace traditional search. Tech executives are betting on the hype first and hoping their talent will make it into a usable product.
- rarx 4y ago
  Even it the chatbot improves, I imagine many folks are waiting for the point where they have some direct control of its training / fine tuning that is easy and accessible. You can correct it, but why waste time doing that if it won’t remember next time? Issues like, e.g. if you have NDAs, are you breaking them by disclosing information with a chatbot? There’s little clarity in regards to those issues, and you can’t just take someone’s word for it. Ideally you’d want control of the chatbot’s learning process and how it handles all kinds of (sensitive?) data.
mrtranscendence 4y ago
That's true enough, but by this point everyone has come to terms with the fact that it can be tough to wade through search results. But Microsoft is presenting the results here as if they were simple, direct, factual statements. That connotes a different feel than search results, I think, and will catch some people off guard as they acclimatize to this new tool. At the very least Microsoft should be doing a better job advertising that the results are frequently incorrect.
- kenjackson 4y ago
  The old school summarized answers that Bing and Google presented had similar issues. The only difference with those is that they had links you could click to see the genesis of the incorrect answer. > At the very least Microsoft should be doing a better job advertising that the results are frequently incorrect. People should realize that information on the internet may or may not be correct in general. Your mileage may vary and you need to decide based on the risk profile associated with the task. If I need to get the start on a poem about two kids at the park, ChatGPT may fit the profile. If I need to know details about how to perform some encryption used for a highly confidential data, then I may go to an authority paper on the topic, skipping things like Wikipedia.
  - mrtranscendence 4y ago
    If "shoulds" were horses we'd all be trampled to death a hundred times over. People should realize that information on the internet may be incorrect, but until everyone does I maintain that Microsoft should do a better job cautioning users that Bing's direct statements of fact may be false.
    - kenjackson 4y ago
      This is what Microsoft says about it on their Bing page, which seems pretty reasonable. I guess you can make the user click through this every time it gives a response, but Wikipedia doesn't make me do that either, nor does Bing/Google for general search results. YouTube doesn't do this when I start a video. --------------------------- Are Bing's AI-generated responses always factual? Bing aims to base all its responses on reliable sources - but AI can make mistakes, and third party content on the internet may not always be accurate or reliable. Bing will sometimes misrepresent the information it finds, and you may see responses that sound convincing but are incomplete, inaccurate, or inappropriate. Use your own judgment and double check the facts before making decisions or taking action based on Bing's responses. To share site feedback or report a concern, open the menu at the top right corner of a response, and then click the flag icon. You can also use the feedback button at the bottom right of every Bing page.
throwuxiytayq 4y ago
> does anyone actually believe these models will be right most of the time? They’re already right most of the time for real world queries, especially compared to web search which these days almost never gives me what I’m looking for at the top of the results (and often even in the first query). People go out of their way to find failure points and then brag online about having outsmarted the AI as if anyone promised that this tech is infallible. It’s quite embarrassing, really.
- mrtranscendence 4y ago
  How is asking Bing about your own name "going out of [your] way to find failure points"?
  - throwuxiytayq 4y ago
    I just googled myself and it’s a pile of nonsense, none of these people are me! slide_whistle.mp3
    - mrtranscendence 4y ago
      Eh. Get back to me when Google purports to be able to understand human language and give factual answers that intelligently integrate search results.
wpietri 4y ago
> does anyone actually believe these models will be right most of the time People tend to have a high level of trust for the stuff that shows up near the top of search. The same goes from products with trusted brands. And note that Google is high on the list: https://morningconsult.com/most-trusted-brands-2020/ https://morningconsult.com/most-trusted-brands-2020/ So if trusted brands want to use this technology, they'll need to do it in ways that either strongly distance it from their brands or that heavily suppress the bad stuff.

vsareto 4y ago

Once they became a serious product pushed by major tech companies, people aren't unreasonable in expecting it to be right most of the time. Especially if it's going to go alongside or replace traditional search. Tech executives are betting on the hype first and hoping their talent will make it into a usable product.

rarx 4y ago
Even it the chatbot improves, I imagine many folks are waiting for the point where they have some direct control of its training / fine tuning that is easy and accessible. You can correct it, but why waste time doing that if it won’t remember next time? Issues like, e.g. if you have NDAs, are you breaking them by disclosing information with a chatbot? There’s little clarity in regards to those issues, and you can’t just take someone’s word for it. Ideally you’d want control of the chatbot’s learning process and how it handles all kinds of (sensitive?) data.

mrtranscendence 4y ago

That's true enough, but by this point everyone has come to terms with the fact that it can be tough to wade through search results. But Microsoft is presenting the results here as if they were simple, direct, factual statements. That connotes a different feel than search results, I think, and will catch some people off guard as they acclimatize to this new tool. At the very least Microsoft should be doing a better job advertising that the results are frequently incorrect.

kenjackson 4y ago
The old school summarized answers that Bing and Google presented had similar issues. The only difference with those is that they had links you could click to see the genesis of the incorrect answer. > At the very least Microsoft should be doing a better job advertising that the results are frequently incorrect. People should realize that information on the internet may or may not be correct in general. Your mileage may vary and you need to decide based on the risk profile associated with the task. If I need to get the start on a poem about two kids at the park, ChatGPT may fit the profile. If I need to know details about how to perform some encryption used for a highly confidential data, then I may go to an authority paper on the topic, skipping things like Wikipedia.
- mrtranscendence 4y ago
  If "shoulds" were horses we'd all be trampled to death a hundred times over. People should realize that information on the internet may be incorrect, but until everyone does I maintain that Microsoft should do a better job cautioning users that Bing's direct statements of fact may be false.
  - kenjackson 4y ago
    This is what Microsoft says about it on their Bing page, which seems pretty reasonable. I guess you can make the user click through this every time it gives a response, but Wikipedia doesn't make me do that either, nor does Bing/Google for general search results. YouTube doesn't do this when I start a video. --------------------------- Are Bing's AI-generated responses always factual? Bing aims to base all its responses on reliable sources - but AI can make mistakes, and third party content on the internet may not always be accurate or reliable. Bing will sometimes misrepresent the information it finds, and you may see responses that sound convincing but are incomplete, inaccurate, or inappropriate. Use your own judgment and double check the facts before making decisions or taking action based on Bing's responses. To share site feedback or report a concern, open the menu at the top right corner of a response, and then click the flag icon. You can also use the feedback button at the bottom right of every Bing page.

throwuxiytayq 4y ago

> does anyone actually believe these models will be right most of the time? They’re already right most of the time for real world queries, especially compared to web search which these days almost never gives me what I’m looking for at the top of the results (and often even in the first query). People go out of their way to find failure points and then brag online about having outsmarted the AI as if anyone promised that this tech is infallible. It’s quite embarrassing, really.

mrtranscendence 4y ago
How is asking Bing about your own name "going out of [your] way to find failure points"?
- throwuxiytayq 4y ago
  I just googled myself and it’s a pile of nonsense, none of these people are me! slide_whistle.mp3
  - mrtranscendence 4y ago
    Eh. Get back to me when Google purports to be able to understand human language and give factual answers that intelligently integrate search results.

wpietri 4y ago

> does anyone actually believe these models will be right most of the time People tend to have a high level of trust for the stuff that shows up near the top of search. The same goes from products with trusted brands. And note that Google is high on the list: https://morningconsult.com/most-trusted-brands-2020/ https://morningconsult.com/most-trusted-brands-2020/ So if trusted brands want to use this technology, they'll need to do it in ways that either strongly distance it from their brands or that heavily suppress the bad stuff.