5 ms·

Major leak reveals new version of Microsoft Bing powered by ChatGPT-4 AI

47 points by jacooper 4y ago

jacooper 4y ago
The search engine revolution has begun. Fuck these companies who kept this technology from us to keep their dumb business undisrupted(Meta and google). Also most legal compaints should be solved now since it highlights the source right?
- flangola7 4y ago
  To be fair, letting SOTA models out into the wild is basically the worst thing you could do when it comes to the control problem.
  - blackflame7000 4y ago
    Can you explain what you mean by control problem?
    - flangola7 4y ago
      As in, how we do control AI so that it aligns with our values? If tomorrow we activate a generalized AI that is able to outthink us, we want that AI to care about things like "do not extinguish all life on Earth." We don't know how far we are from building such an AI. The only somewhat safe strategy is to go slow and be extremely methodical and diligent in our research, and that goes out the window if you start open sourcing it and letting anyone poke at it. Even if we never reach generalized super-intelligent AI we don't want some guy in a basement having access to an AI that can carry out the command "hack into the electrical grids of all the countries I don't like and burn out their core transformers." I'm not worried about AGI (yet) but I believe some kind of limited-scope but extremely lethal AI like the hacking scenario I just described will be possible very soon, possibly in a matter of months or low single-digit years. In the near future we could see global proliferation controls on AI research and computing hardware in the same way there are controls on uranium mining and high-speed centrifuges. https://old.reddit.com/r/ControlProblem/comments/za3haw/welcome_start_here https://old.reddit.com/r/ControlProblem/comments/za3haw/welc...
      - blackflame7000 4y ago
        Isn’t the goal of OpenAI to be open to everyone including that guy in the basement (lets hope he’s at the NSA hq in this timeline)
- epistemer 4y ago
  I expect to be bypassing the search engine none too soon. Up to date chatGPT that could both summarize and link to the content would pretty much end my use of a search engine. That will just become a natural part of the back and forth with the AI assistant. I suspect we massively overrate the first mover advantage with new technology like this though. I remember not being able to join Friendster because I didn't know anyone on Friendster to invite me. An invite only social network is obviously stupid now but that was hardly obvious in 2003. chatGPT will probably make a huge fatal mistake exactly because they are first. I wouldn't count Google and Meta out so easily.
  - jacooper 4y ago
    > Up to date chatGPT that could both summarize and link to the content would pretty much end my use of a search engine. That will just become a natural part of the back and forth with the AI assistant. This what the leak says Bing is planning to do. > I wouldn't count Google and Meta out so easily For sure, Google has a better model already. But I meant that without ChatGPT and openAI we would've been still stuck with crappy search engines.
ddmma 4y ago
Open innovation aligned to the current global state of mind. Congratulations on the development of new human capabilities using the latest technologies. Happy moments to be alive an explore further the unknowns
- ddmma 4y ago
  This was generated in Bing :(
lagrange77 4y ago
This is a good opportunity for MS in the search field and i'm interested in what Google has to compete. But i have also a bad gut feeling about this happening too early.
mshake2 4y ago
"Major leak" aka guerrilla marketing. C'mon, are we still pretending that these new product "leaks" are unplanned? Just preview the damn thing instead of acting like we're subverting The Man by viewing their marketing.
- aliqot 4y ago
  if you preview and there's a bug or another Tay incident, you can't say "this wasn't production ready", if it's a leak you have plausible deniability to save face.
- NoPicklez 4y ago
  C'mon, why are we still commenting on this at all. Don't get hung up on the wording.
- brookst 4y ago
  Have you ever worked in product management or marketing? Leaks are not desirable. On the rare occasions when the company wants press before announcement, they talk to a few of their pet journalists who they can trust to frame things appropriately. Random leaks from random people who don’t use the right language is a nightmare, and risks rolling the dice on the narrative.
ecshafer 4y ago
I don't know why I would want to query a ML algorithm to generate an answer as opposed to finding a webpage to read about the topic. Sure sometimes I am just looking for "Who was the actor in Alien?" and want a name to spit out. But most of the time I am using the query as a key to a larger amount of information. Search has gotten worse over time, but I don't think this solves that problem. Instead I think the corpus for these will end up being corrupted by SEO and we will have garbage there as well. Increasingly I want a search that is just Google 20 years ago, Kagi is the most promising company in this space imo.
- Always42 4y ago
  1 possibility, not to be plummeted by ads.
  - shostack 4y ago
    Which is ironic because that's how this will likely end... In MS serving ads based on your queries and embedded within. Imagine if you asked it what to take for a headache and it embedded a pharma brand with the most minimal ad disclosure they can get their lawyers to agree to. At least with display ads it is blindingly obvious it's an ad.
    - genericone 4y ago
      Microsoft's cost-per-click for ads embedded in a chatgpt query/response is going to be an order of magnitude higher, since the engine will be able to reason linguistically about what ads are best.
- NoPicklez 4y ago
  You might not always want to, but sometimes you might. Sometimes you don't want to read about it, sometimes you just want an answer. The great thing about search is that is covered a wide range of needs.
- brundolf 4y ago
  I know people who have struggled to learn eg Python programming just from googling (because they lacked any programming background at all so they didn't even know what to search for to have the foundational stuff explained), and then tried using ChatGPT instead and made tons of progress because it could figure out what they meant to ask, and then digest and tune the answers to meet them where they were at Of course it's sometimes wrong, so it's good to verify, which is why it's good that the Bing version cites its sources
CommanderData 4y ago
I find myself reaching for Chatgpt often when I realised I'd save time getting my answer generated for me instead of reading 4-5 sites to try and find what I wanted. It's a very different experience.
bandyaboot 4y ago
For when you have a question and simply want an answer that sounds like it could be right.
jrnichols 4y ago
I'm not super familiar with chatGPT but how well does this address the issue of outdated blog posts and ancient reddit/forum posts?
- anaganisk 4y ago
  Outdated? The other I asked it to cite its sources, it generated real looking links to 404 pages. It just built fake links, not even old links.
  - blackflame7000 4y ago
    I guess the sources were anonymous but that’s standard for modern journalism
- jacooper 4y ago
  Its supposedly going to generate summaries from the search results, like how https://www.perplexity.ai https://www.perplexity.ai works.
tracerbulletx 4y ago
From the article it sounds like a few people got served a test and they made it out to be some kind of national security incident.
freediver 4y ago
I hope we also get an answer to the question who will be paying for the information served - users or advertisers?
blackflame7000 4y ago
[flagged]
slimebot80 4y ago
Can't wait for my future science and medical students to confidently learn from answers that are good enough to be convincing.
brundolf 4y ago
Would this apply to DuckDuckGo too, since it's powered by Bing?
- jacooper 4y ago
  Doubt Bing is going to give its ai search to DDG anytime soon.

47 points by jacooper 4y ago

The search engine revolution has begun. Fuck these companies who kept this technology from us to keep their dumb business undisrupted(Meta and google). Also most legal compaints should be solved now since it highlights the source right?

flangola7 4y ago
To be fair, letting SOTA models out into the wild is basically the worst thing you could do when it comes to the control problem.
- blackflame7000 4y ago
  Can you explain what you mean by control problem?
  - flangola7 4y ago
    As in, how we do control AI so that it aligns with our values? If tomorrow we activate a generalized AI that is able to outthink us, we want that AI to care about things like "do not extinguish all life on Earth." We don't know how far we are from building such an AI. The only somewhat safe strategy is to go slow and be extremely methodical and diligent in our research, and that goes out the window if you start open sourcing it and letting anyone poke at it. Even if we never reach generalized super-intelligent AI we don't want some guy in a basement having access to an AI that can carry out the command "hack into the electrical grids of all the countries I don't like and burn out their core transformers." I'm not worried about AGI (yet) but I believe some kind of limited-scope but extremely lethal AI like the hacking scenario I just described will be possible very soon, possibly in a matter of months or low single-digit years. In the near future we could see global proliferation controls on AI research and computing hardware in the same way there are controls on uranium mining and high-speed centrifuges. https://old.reddit.com/r/ControlProblem/comments/za3haw/welcome_start_here https://old.reddit.com/r/ControlProblem/comments/za3haw/welc...
    - blackflame7000 4y ago
      Isn’t the goal of OpenAI to be open to everyone including that guy in the basement (lets hope he’s at the NSA hq in this timeline)
epistemer 4y ago
I expect to be bypassing the search engine none too soon. Up to date chatGPT that could both summarize and link to the content would pretty much end my use of a search engine. That will just become a natural part of the back and forth with the AI assistant. I suspect we massively overrate the first mover advantage with new technology like this though. I remember not being able to join Friendster because I didn't know anyone on Friendster to invite me. An invite only social network is obviously stupid now but that was hardly obvious in 2003. chatGPT will probably make a huge fatal mistake exactly because they are first. I wouldn't count Google and Meta out so easily.
- jacooper 4y ago
  > Up to date chatGPT that could both summarize and link to the content would pretty much end my use of a search engine. That will just become a natural part of the back and forth with the AI assistant. This what the leak says Bing is planning to do. > I wouldn't count Google and Meta out so easily For sure, Google has a better model already. But I meant that without ChatGPT and openAI we would've been still stuck with crappy search engines.

ddmma 4y ago

Open innovation aligned to the current global state of mind. Congratulations on the development of new human capabilities using the latest technologies. Happy moments to be alive an explore further the unknowns

ddmma 4y ago
This was generated in Bing :(

lagrange77 4y ago

This is a good opportunity for MS in the search field and i'm interested in what Google has to compete. But i have also a bad gut feeling about this happening too early.

mshake2 4y ago

"Major leak" aka guerrilla marketing. C'mon, are we still pretending that these new product "leaks" are unplanned? Just preview the damn thing instead of acting like we're subverting The Man by viewing their marketing.

aliqot 4y ago
if you preview and there's a bug or another Tay incident, you can't say "this wasn't production ready", if it's a leak you have plausible deniability to save face.
NoPicklez 4y ago
C'mon, why are we still commenting on this at all. Don't get hung up on the wording.
brookst 4y ago
Have you ever worked in product management or marketing? Leaks are not desirable. On the rare occasions when the company wants press before announcement, they talk to a few of their pet journalists who they can trust to frame things appropriately. Random leaks from random people who don’t use the right language is a nightmare, and risks rolling the dice on the narrative.

ecshafer 4y ago

I don't know why I would want to query a ML algorithm to generate an answer as opposed to finding a webpage to read about the topic. Sure sometimes I am just looking for "Who was the actor in Alien?" and want a name to spit out. But most of the time I am using the query as a key to a larger amount of information. Search has gotten worse over time, but I don't think this solves that problem. Instead I think the corpus for these will end up being corrupted by SEO and we will have garbage there as well. Increasingly I want a search that is just Google 20 years ago, Kagi is the most promising company in this space imo.

Always42 4y ago
1 possibility, not to be plummeted by ads.
- shostack 4y ago
  Which is ironic because that's how this will likely end... In MS serving ads based on your queries and embedded within. Imagine if you asked it what to take for a headache and it embedded a pharma brand with the most minimal ad disclosure they can get their lawyers to agree to. At least with display ads it is blindingly obvious it's an ad.
  - genericone 4y ago
    Microsoft's cost-per-click for ads embedded in a chatgpt query/response is going to be an order of magnitude higher, since the engine will be able to reason linguistically about what ads are best.
NoPicklez 4y ago
You might not always want to, but sometimes you might. Sometimes you don't want to read about it, sometimes you just want an answer. The great thing about search is that is covered a wide range of needs.
brundolf 4y ago
I know people who have struggled to learn eg Python programming just from googling (because they lacked any programming background at all so they didn't even know what to search for to have the foundational stuff explained), and then tried using ChatGPT instead and made tons of progress because it could figure out what they meant to ask, and then digest and tune the answers to meet them where they were at Of course it's sometimes wrong, so it's good to verify, which is why it's good that the Bing version cites its sources

CommanderData 4y ago

I find myself reaching for Chatgpt often when I realised I'd save time getting my answer generated for me instead of reading 4-5 sites to try and find what I wanted. It's a very different experience.

bandyaboot 4y ago

For when you have a question and simply want an answer that sounds like it could be right.

jrnichols 4y ago

I'm not super familiar with chatGPT but how well does this address the issue of outdated blog posts and ancient reddit/forum posts?

anaganisk 4y ago
Outdated? The other I asked it to cite its sources, it generated real looking links to 404 pages. It just built fake links, not even old links.
- blackflame7000 4y ago
  I guess the sources were anonymous but that’s standard for modern journalism
jacooper 4y ago
Its supposedly going to generate summaries from the search results, like how https://www.perplexity.ai https://www.perplexity.ai works.

tracerbulletx 4y ago

From the article it sounds like a few people got served a test and they made it out to be some kind of national security incident.

freediver 4y ago

I hope we also get an answer to the question who will be paying for the information served - users or advertisers?

blackflame7000 4y ago

[flagged]

slimebot80 4y ago

Can't wait for my future science and medical students to confidently learn from answers that are good enough to be convincing.

brundolf 4y ago

Would this apply to DuckDuckGo too, since it's powered by Bing?

jacooper 4y ago
Doubt Bing is going to give its ai search to DDG anytime soon.