10 ms·

Nobody seems to be bringing up the questions inherent to a big company moving on this that will have to follow the "rules": - Why would I ever let bing crawl m

Nobody seems to be bringing up the questions inherent to a big company moving on this that will have to follow the "rules": - Why would I ever let bing crawl my site if they aren't going to send any visitors to me? - fair use of snippets has relied on them being brief and linking to the source. Lawsuits will be immediate. I do love these imaginary scenarios where ChatGPT is going to find me the best air fryer, though. Where is that information going to come from, exactly? Barely anyone is making money writing reviews today, most sites are farmed content. What happens when even the ok sites' reviews are quickly scraped and put into the next model iteration? Bing is going to have to come up with some kind of radical revenue sharing too if they want anything written after 2023.

darawk 4y ago
> Why would I ever let bing crawl my site if they aren't going to send any visitors to me? I don't think it's up to you, legally speaking: https://en.wikipedia.org/wiki/HiQ_Labs_v._LinkedIn https://en.wikipedia.org/wiki/HiQ_Labs_v._LinkedIn I mean, they could be nice and respect your robots.txt, but they certainly don't have to. > fair use of snippets has relied on them being brief and linking to the source. Lawsuits will be immediate. It's possible that fair use law will be expanded to cover this case, but as constructed the output of these models is generally fairly derivative of any specific original, and so probably protected under fair use. If it were spitting out exact copies of things it had read, it would probably be pretty easy to train that behavior out of it. > I do love these imaginary scenarios where ChatGPT is going to find me the best air fryer, though. Where is that information going to come from, exactly? Barely anyone is making money writing reviews today, it's mostly farmed content. What happens when even those sites' reviews are quickly scraped and put into the next model iteration? Bing is going to have to come up with some kind of radical revenue sharing too if they want anything fresh. I do agree with this, though. The LLMification of search is going to squeeze revenue for content creators of all kinds to literally nothing, at least if that content isn't paywalled. Which probably means that that's exactly where we're headed.
- magicalist 4y ago
  > I don't think it's up to you, legally speaking: https://en.wikipedia.org/wiki/HiQ_Labs_v._LinkedIn https://en.wikipedia.org/wiki/HiQ_Labs_v._LinkedIn > I mean, they could be nice and respect your robots.txt, but they certainly don't have to. That case was limited to the CFAA, but you seem to get the gist of what I'm saying when I specified it's different when it's Microsoft doing the scraping. If Bing starts ignoring robots.txt and data still start showing up in their results, all the early 2000s lawsuits are going to be opened back up. > It's possible that fair use law will be expanded to cover this case, but as constructed the output of these models is generally fairly derivative of any specific original, and so probably protected under fair use. Unless there's a reason for them to be considered fair use, derivative works are going to lose a copyright suit. And what's the fair use argument? If I'm the only one on the internet saying something and suddenly ChatGPT can talk about the same thing and I'm losing money as a result, there's no fair use argument there. Search engines won those early lawsuits by being transformative (index vs content), minimal, and linking to their source. None of that would apply here.
  - jaspax 4y ago
    What GP means is that ChatGPT output is generally not similar enough to any _particular_ source document to establish the fact that it's derivative. Instead, it resembles what you'd get if you asked a (credulous and slightly dumb) human to read a selection of documents and then summarize them. These kinds of summaries are absolutely not copyright violations, even if the source document can actually be identified.
    - Xelynega 4y ago
      > ChatGPT output is generally not similar enough to any _particular_ source document to establish the fact that it's derivative. Isn't this exactly what a court case would be trying to clarify? If so wouldn't assuming this be begging the question?
- toteno 4y ago
  Sadly, seem like the decision in that case was changed. From your link: > In a November 2022 ruling the Ninth Circuit ruled that hiQ had breached LinkedIn's User Agreement and a settlement agreement was reached between the two parties.
neel8986 4y ago
This is an extremely important point. Something like ChatGPT without attribution can completely kill the open web. Every company will keep their information in closed walled garden if no traffic is flowing through them . I don't see a scenario where something like stackoverflow can exist if no one goes to the site.
- imhoguy 4y ago
  I think StackOverflow will exist and do well. 1st, it is source of information for ChatGPT itself so if there would be no new content then AI is going to implode too. 2nd, very often I skip top answer because it has some edge cases or simply is outdated. The answer comments often highlight such issues. I don't think ChatGPT could be trusted without verification, not in serious programming work.
  - emoII 4y ago
    On the first point, that is no guarantee that users will stay on the site. The AI is currently only using data from 2021 and earlier as far as I'm aware, and does so without feeling out of date. Before we see any significant signs of the AI imploding due to lack of new information, SO might well be long gone
  - phpisthebest 4y ago
    I see Stack overflow as one of the problems here. StackOverflow went along way in killing the Tech Blog, and the number of "right" but poor answers on Stack sites are at an all time high Often the "best" answer on those sites is buried or even downvoted in favor of an answer that "just works" but may have security issues, maintainability issues, is out dated, etc. In alot of area's I find Stack answer to be of low quality if you happen to have any indepth knowelege of that area.
    - OOPMan 4y ago
      Indeed. They should be renamed to ShitOverflow, because that's how bad the quality is a lot of the time.
hankchinaski 4y ago
>Why would I ever let bing crawl my site if they aren't going to send any visitors to me? Google already does this with featured snippets
- sct202 4y ago
  Which were already highly unpopular with websites, but at least have some attribution.
  - brookst 4y ago
    So we’re speculating that the Bing chatGPT implementation will crawl public websites, answer queries strictly from its training data, and present unattributed snippets? That does sound both flawed as a search engine and objectionable to site operators. In addition to not being announced or even rumored to work that way. So, maybe the implementation is different from that model?
hooande 4y ago
> "Bing is going to have to come up with some kind of radical revenue sharing too if they want anything written after 2023." ChatGPT doesn't include anything written after 2021. I certainly wouldn't use it to find an air fryer. The results will be from over a year ago. I would want to see what the newest air fryer options are and it would be really important to have to up to date pricing. AFAIK there is not a way to update a large language model in real time. You have to train on the entire dataset to do a meaningful update, just like with most forms of neural networks. For ChatGPT that takes days and costs hundreds of thousands of dollars. Every time. It's great for explanations of concepts, and programming, and a few other things. But with the huge caveat that all of the information you're looking at is from one year ago and may have changed in that time. This really limits the utility of ChatGPT for me.
- wepple 4y ago
  That’s a very solvable problem though. If Microsoft decides to integrate ChatGPT with bing, they have the resources to retrain the model on a more recent data set, and even do it somewhat regularly
  - typon 4y ago
    You don't even need to retrain if you use retrieval transformers. That is the real revolution waiting to happen. Deepmind already unlocked it with RETRO, but I don't know why a public version hasn't been released - hooked into the live internet.
    - IanCal 4y ago
      OpenAI have webgpt too https://openai.com/blog/webgpt/ https://openai.com/blog/webgpt/
- jacooper 4y ago
  OpenAI is already working on solving this https://openai.com/blog/webgpt/ https://openai.com/blog/webgpt/
asah 4y ago
Yeah, not helpful. But long winded. There are many air fryers on the market and the best one for you will depend on your needs and preferences. Some factors to consider when selecting an air fryer include size, price, features, and overall performance. Some popular air fryers to consider include the Philips Airfryer, the Ninja Foodi, and the Cosori Air Fryer. It might be helpful to read online reviews and compare the features of different models to find the one that works best for you.
jackbrookes 4y ago
Websites will optimise for AI eyes rather than Human eyes. Advertisers will pay to embed information in the websites that is read by AI, which subtlety makes the advertisers' products more valuable in the eyes of the AI. Then the AI would ultimately spit out information to users that are biased towards the advertiser's products.
- woeirua 4y ago
  We already know that advertisers aren’t willing to pay that much for “subliminal” advertising. People have been trying to do product placement in movies and shows forever and it’s never really taken off.
  - shkkmo 4y ago
    This seems factually incorrect. It's hard to find consistent historical numbers but what I can find implies pretty steady double digit growth over the last decade or two. If you have good sources that say otherwise, I'd love to see them.
  - brookst 4y ago
    Product placement is everywhere. Next time you watch a movie or show, look for the clothing brands, computer brands, car brands, wine brands, etc. everywhere. And think about sponsorships. From soccer to nascar, sports is covered with branding.
    - Spivak 4y ago
      That's not subliminal, you're describing sponsorships (i.e. manufactured social proof).
      - brookst 4y ago
        "Subliminal" and "sponsorship" are totally orthogonal. One refers to the presentation, the other the business arrangement.
  - 4y ago
cmrdporcupine 4y ago
My guess is your first point is exactly why Google hasn't done this yet. Their 'knowledge boxes' are already crossing a line that in general they felt nervous about crossing historically, but they don't go very far. Google on the whole historically did not want to alienate publishers (and the advertisers that hang out on publisher content) and has avoided being in the content production business for this reason.
wongarsu 4y ago
If language models take over text content, content creators will flee even quicker into creating video content. There's already a trend where younger people tend to prefer video for being more "genuine", and now it might become a sign of "human made" for a couple years. Also easier to monetize, and easier to build parasocial relationships, so all around a plus for creators. Too bad I prefer text.
- ikt 4y ago
  could it somehow get access to the subtitles and then use them to answer queries? also i hope this comes to ecosia, would like to experiment and try it at least
  - jadbox 4y ago
    Exactly, and many bots exist today to mine user videos for the automated subtitle information. In other words, there's no escaping GPT from learning from any kind of medium.
  - throwaway09223 4y ago
    > " could it somehow get access to the subtitles and then use them to answer queries?" It's not even necessary - computers are already excellent at understanding spoken words. Have you tried automatic captioning recently? Half the inputs to my phone are already voice, not text. Video is a harder problem, but it's not too far behind.
- belltaco 4y ago
  I prefer text too but I feel like that's mostly because the videos are not information dense on purpose. They expand to whatever the youtube algorithm prefers at the time, which is about 10 minutes now. Ironically, tiktoks are more information dense but the search is completely useless.
  - bronco21016 4y ago
    I’m finding more and more that the information density isn’t there because the video content is actually just an infomercial for a “course”.
WhiteNoiz3 4y ago
IMO this is the big problem with the internet as it exists today - there is no incentive for producing accurate, unbiased information and non-sensationalist opinions. My greatest hope for the future is that somehow we can incentivize people to produce "good" information for AI based assistants and move away from the rage/shock based advertising model that most of the internet currently uses. Personally I would rather pay a few cents for a query that produces valuable results and doesn't put me in a bad mood than pay with my time and attention like we do today. AI systems will absolutely need to be able to identify the training sources with every result (even if it is coming from several sources) and those sources should be compensated. IMO that's the only fair model for both image and text generation that is based on authors and artists work.
- andsoitis 4y ago
  > problem with the internet as it exists today - there is no incentive for producing accurate, unbiased information and non-sensationalist opinions. I think this problem is orthogonal to the internet as medium, though I’ll concede that it has proven to be the biggest amplifier of this dynamic. Correct (or correct as far as humans know, or most likely correct, etc.) costs money to create. False or completely made up information costs nothing, plus has the potential upside of sensationalism, thus further increasing its ROI. Agree with your point about developing more incentives for correct information and penalties for false.
- int_19h 4y ago
  It's not just that there's no incentive for that, but there's a very strong incentive to do the exact opposite: https://www.youtube.com/watch?v=rE3j_RHkqJc https://www.youtube.com/watch?v=rE3j_RHkqJc
dalbasal 4y ago
These questions are constant. I do think you bring up relevant issues, but they aren't quite showstoppers. Websites allow SE crawlers because (a) whatever traffic they get is better than not traffic (b) because allowing crawlers is default and doesn't cost anything and (c) google/bing don't negotiate. They are one, sites are many. This has already played out in news. News outlets wanted Google to pay for content. Google (initially) responded by allowing them to opt out of Google. Over the years, they have negotiated a little bit. Courts, in some places, forced Google to negotiate... It's news and politicians care about news specifically. Overall though, there have not been meaningful moments where people got pissed off with Google and blocked crawlers. Not newspapers and not anyone else. Site owners being mad doesn't affect google or Bing. What does matter to search engines is walled gardens. Facebook pioneered this, and this does matter to Google. There is, in a lot of cases, a lot less content to index and serve users. All those old forums, for example. These are search problems, and GPT-based search will inherit them. ChatGPT will have the same problem recommending the best air fryer as normal search does. GPT is a different way of presenting information... it's not presenting different information. RE: Lawsuits. Again, history. Youtube, for example, started off with rampant copyright infringement. But, legal systems were primitive. Lawyers and legislatures didn't know what to do. Claimants were extremely dispersed, and would have had to pioneer case law. Ultimately, copyright took >10 years to really apply online and by that point youtube and other social media was entrenched. The law lags. In practice, early movers are free to operate flawlessly and they get to shut the door after them. Now that Google is firmly entrenched, copyright law serves as one of their trenches.
- jredwards 4y ago
  Incidentally, law seems like an incredibly powerful potential application for ChatGPT.
VLM 4y ago
> Where is that information going to come from, exactly? Manufacturers, with quality ranging from excellent to trash. Consider trying to buy a 1K resistor at Digikey using their parametric search. Possible, but tedious and time consuming because you need a lot of domain knowledge to know what you want, and the technological range of "things with 1K of resistance" is extremely vast. At least its possible because the mfgrs are honest when Digikey imports their data. Consider the opposite, consumer goods. 500 watt PC power supplies with random marketing number stickers on the same chassis ranging from 500 to 1200 watts. Consumer level air compressors and consumer level vacuum cleaners than plug into household wall outlets claiming "8 horsepower" or whatever insane marketing nonsense. Clothes with vanity sizing so a "medium" tag fits like a real world XXL. Every processed food in a store with a "keto" label is high carb sugar added garbage, much like happened with "organic" label in the old days (the employees at the farm, distributor, warehouse, and/or retail store level take the same produce out of one bin and put it in two places with different prices) I think it will help when purchasing technical engineering type products but be an epic fail at inherently misleading consumer goods.
- Dwolb 4y ago
  Maybe manufacturers could upload their design docs and ChatGPT could learn exactly what the object does and what its performance parameters are.
- Xelynega 4y ago
  If you're trying to search for a specific resistor without the prerequisite domain knowledge, how will you be able to vet whether or not the answer given by a language model meets your needs? Imagining that language models like gpt will ever be able to index up-to-date information is literally trying to apply the concept of "artificial intelligence" to a probabilistic language model. It's incompatible with what it's actually doing.
ctoth 4y ago
Oh gosh, maybe we'll actually have to pay for things, and we'll find that the market for the fifth random blog trying to make money off of free information using ads doesn't really exist. What a terrible world this will obviously be. No. The weird thing is this idea that because you put ads on your site, you deserve money. Your ads are making the Internet worse. You probably don't realize this, because you most-likely use an ad blocker, which means you want people too dumb to use ad blockers to subsidize the web that you can use for free, but the current web is working well for approximately no one. Would I pay $5 a month for StackOverflow if it didn't show up for everything I Google? Most likely. Would this be a better world? almost certainly. We tried the thing with ads. It sucks. I welcome our new AI search overlords.
- skinnymuch 4y ago
  Why would you want power centralized? Big corporations are never your friend.
  - int_19h 4y ago
    Power is also centralized when most supposedly independent actors buy ads from the same large advertisers, and utterly depend on their income from those ads to do whatever they're doing.
sieabahlpark 4y ago
[dead]
schaefer 4y ago
How I’d like to weather this storm: 1) Everyone cryptographically signs their work for identity confirmation. 2) there exists a blockchain whose sole purpose is to allow content creators to establish copyright date on a digital piece of work. 3) a public that uses the two items above what evaluating the reputation of an artist.
- p0pcult 4y ago
  This seems to make a lot of sense. The artists themselves also have an incentive to be blockchain validators/miners, thereby reducing the need for token payout, and the subsequent speculation that comes with tokenization (I think).
- codethief 4y ago
  How does that prevent anyone from using ChatGPT to generate new (supposedly human-written) content?
  - p0pcult 4y ago
    It doesn't; however, the signature in your hypothetical doesn't correspond to a known/trusted author.
- marwis 4y ago
  You don't need blockchain for cryptographic timestamp
  - schaefer 4y ago
    I've got some reading to do[1]. Thank you for the head's up. [1] https://en.wikipedia.org/wiki/Trusted_timestamping https://en.wikipedia.org/wiki/Trusted_timestamping
rmbyrro 4y ago
Their plan is to use Neuralink to pull all the information from people's brains.
_huayra_ 4y ago
> Barely anyone is making money writing reviews today, most sites are farmed content. I'm sure ChatGPT will be able to write a bunch of terrible SEO prose that precedes the actual air fryer review (or worse, recipe) about how the author's grandma had an air fryer when she was young and remembered the great times with her grandma (etc), for roughly 95% of the text! In all seriousness, being able to swerve all that terrible SEO content on reviews will always be welcome!
waynesonfire 4y ago
> Why would I ever let bing crawl my site if they aren't going to send any visitors to me? You tell me! It's your site. If you want money maybe you should charge for your content? And honestly, the web that Google presents is just so terrible that I don't want to visit your site, unfortunately. And, maybe it's a price worth paying.
throwaway4aday 4y ago
A language model that provides answers with sources (which could be found using a traditional search engine that searches the corpus that the language model is trained on) would be very useful and would also allow it to link directly to the source material. The trouble would be in finding the exact sources since the output of the language model is unlikely to be verbatim but current search engines can deal with imprecise queries fairly well so it's not an intractable problem. A very well curated data set would help this immensely. I'd be super interested in a language model that was able to synthesize knowledge drawn from a large corpus of books and then cite relevant sections from various titles.
bioemerl 4y ago
> Why would I ever let bing crawl my site if they aren't going to send any visitors to me? How can you refuse? The only way I know would be to require an account, but even then they could bypass it.
- rafale 4y ago
  Major search engines honor robots.txt
  - bioemerl 4y ago
    If it becomes standard to have such a file and it effects their bottom line, could they disregard it?
    - Xelynega 4y ago
      If the file that were previously honored as consent to use the copyright material is subsequently ignored, wouldn't the content creators take the indexers to court for copyright infringement?
danielrhodes 4y ago
I don't see why this would be a copyright violation anymore than somebody learning something from multiple sources and reformulating what they learned into an answer to a question. As long as it isn't explicitly reciting its training data, there shouldn't be an issue of copyright.
sebzim4500 4y ago
>- Why would I ever let bing crawl my site if they aren't going to send any visitors to me? They will send at least some visitors, which is better than the zero visitors you will get from bing if you block it. >- fair use of snippets has relied on them being brief and linking to the source. Lawsuits will be immediate. Yes and microsoft has lawyers, who have presumably determined that the cost of fighting these frivolous lawsuits is not overwhelming.
6gvONxR4sf7o 4y ago
Put SEO into the picture and things get hairier. Incredibly realistic spam is about to go through the roof, so search engines will have an insanely harder time distinguishing between useful content and spam. Making money from search traffic to your (presumably useful) site is going to get harder in a bunch of ways, due to generative models.

darawk 4y ago

> Why would I ever let bing crawl my site if they aren't going to send any visitors to me? I don't think it's up to you, legally speaking: https://en.wikipedia.org/wiki/HiQ_Labs_v._LinkedIn https://en.wikipedia.org/wiki/HiQ_Labs_v._LinkedIn I mean, they could be nice and respect your robots.txt, but they certainly don't have to. > fair use of snippets has relied on them being brief and linking to the source. Lawsuits will be immediate. It's possible that fair use law will be expanded to cover this case, but as constructed the output of these models is generally fairly derivative of any specific original, and so probably protected under fair use. If it were spitting out exact copies of things it had read, it would probably be pretty easy to train that behavior out of it. > I do love these imaginary scenarios where ChatGPT is going to find me the best air fryer, though. Where is that information going to come from, exactly? Barely anyone is making money writing reviews today, it's mostly farmed content. What happens when even those sites' reviews are quickly scraped and put into the next model iteration? Bing is going to have to come up with some kind of radical revenue sharing too if they want anything fresh. I do agree with this, though. The LLMification of search is going to squeeze revenue for content creators of all kinds to literally nothing, at least if that content isn't paywalled. Which probably means that that's exactly where we're headed.

magicalist 4y ago
> I don't think it's up to you, legally speaking: https://en.wikipedia.org/wiki/HiQ_Labs_v._LinkedIn https://en.wikipedia.org/wiki/HiQ_Labs_v._LinkedIn > I mean, they could be nice and respect your robots.txt, but they certainly don't have to. That case was limited to the CFAA, but you seem to get the gist of what I'm saying when I specified it's different when it's Microsoft doing the scraping. If Bing starts ignoring robots.txt and data still start showing up in their results, all the early 2000s lawsuits are going to be opened back up. > It's possible that fair use law will be expanded to cover this case, but as constructed the output of these models is generally fairly derivative of any specific original, and so probably protected under fair use. Unless there's a reason for them to be considered fair use, derivative works are going to lose a copyright suit. And what's the fair use argument? If I'm the only one on the internet saying something and suddenly ChatGPT can talk about the same thing and I'm losing money as a result, there's no fair use argument there. Search engines won those early lawsuits by being transformative (index vs content), minimal, and linking to their source. None of that would apply here.
- jaspax 4y ago
  What GP means is that ChatGPT output is generally not similar enough to any _particular_ source document to establish the fact that it's derivative. Instead, it resembles what you'd get if you asked a (credulous and slightly dumb) human to read a selection of documents and then summarize them. These kinds of summaries are absolutely not copyright violations, even if the source document can actually be identified.
  - Xelynega 4y ago
    > ChatGPT output is generally not similar enough to any _particular_ source document to establish the fact that it's derivative. Isn't this exactly what a court case would be trying to clarify? If so wouldn't assuming this be begging the question?
toteno 4y ago
Sadly, seem like the decision in that case was changed. From your link: > In a November 2022 ruling the Ninth Circuit ruled that hiQ had breached LinkedIn's User Agreement and a settlement agreement was reached between the two parties.

neel8986 4y ago

This is an extremely important point. Something like ChatGPT without attribution can completely kill the open web. Every company will keep their information in closed walled garden if no traffic is flowing through them . I don't see a scenario where something like stackoverflow can exist if no one goes to the site.

imhoguy 4y ago
I think StackOverflow will exist and do well. 1st, it is source of information for ChatGPT itself so if there would be no new content then AI is going to implode too. 2nd, very often I skip top answer because it has some edge cases or simply is outdated. The answer comments often highlight such issues. I don't think ChatGPT could be trusted without verification, not in serious programming work.
- emoII 4y ago
  On the first point, that is no guarantee that users will stay on the site. The AI is currently only using data from 2021 and earlier as far as I'm aware, and does so without feeling out of date. Before we see any significant signs of the AI imploding due to lack of new information, SO might well be long gone
- phpisthebest 4y ago
  I see Stack overflow as one of the problems here. StackOverflow went along way in killing the Tech Blog, and the number of "right" but poor answers on Stack sites are at an all time high Often the "best" answer on those sites is buried or even downvoted in favor of an answer that "just works" but may have security issues, maintainability issues, is out dated, etc. In alot of area's I find Stack answer to be of low quality if you happen to have any indepth knowelege of that area.
  - OOPMan 4y ago
    Indeed. They should be renamed to ShitOverflow, because that's how bad the quality is a lot of the time.

hankchinaski 4y ago

>Why would I ever let bing crawl my site if they aren't going to send any visitors to me? Google already does this with featured snippets

sct202 4y ago
Which were already highly unpopular with websites, but at least have some attribution.
- brookst 4y ago
  So we’re speculating that the Bing chatGPT implementation will crawl public websites, answer queries strictly from its training data, and present unattributed snippets? That does sound both flawed as a search engine and objectionable to site operators. In addition to not being announced or even rumored to work that way. So, maybe the implementation is different from that model?

hooande 4y ago

> "Bing is going to have to come up with some kind of radical revenue sharing too if they want anything written after 2023." ChatGPT doesn't include anything written after 2021. I certainly wouldn't use it to find an air fryer. The results will be from over a year ago. I would want to see what the newest air fryer options are and it would be really important to have to up to date pricing. AFAIK there is not a way to update a large language model in real time. You have to train on the entire dataset to do a meaningful update, just like with most forms of neural networks. For ChatGPT that takes days and costs hundreds of thousands of dollars. Every time. It's great for explanations of concepts, and programming, and a few other things. But with the huge caveat that all of the information you're looking at is from one year ago and may have changed in that time. This really limits the utility of ChatGPT for me.

wepple 4y ago
That’s a very solvable problem though. If Microsoft decides to integrate ChatGPT with bing, they have the resources to retrain the model on a more recent data set, and even do it somewhat regularly
- typon 4y ago
  You don't even need to retrain if you use retrieval transformers. That is the real revolution waiting to happen. Deepmind already unlocked it with RETRO, but I don't know why a public version hasn't been released - hooked into the live internet.
  - IanCal 4y ago
    OpenAI have webgpt too https://openai.com/blog/webgpt/ https://openai.com/blog/webgpt/
jacooper 4y ago
OpenAI is already working on solving this https://openai.com/blog/webgpt/ https://openai.com/blog/webgpt/

asah 4y ago

Yeah, not helpful. But long winded. There are many air fryers on the market and the best one for you will depend on your needs and preferences. Some factors to consider when selecting an air fryer include size, price, features, and overall performance. Some popular air fryers to consider include the Philips Airfryer, the Ninja Foodi, and the Cosori Air Fryer. It might be helpful to read online reviews and compare the features of different models to find the one that works best for you.

jackbrookes 4y ago

Websites will optimise for AI eyes rather than Human eyes. Advertisers will pay to embed information in the websites that is read by AI, which subtlety makes the advertisers' products more valuable in the eyes of the AI. Then the AI would ultimately spit out information to users that are biased towards the advertiser's products.

woeirua 4y ago
We already know that advertisers aren’t willing to pay that much for “subliminal” advertising. People have been trying to do product placement in movies and shows forever and it’s never really taken off.
- shkkmo 4y ago
  This seems factually incorrect. It's hard to find consistent historical numbers but what I can find implies pretty steady double digit growth over the last decade or two. If you have good sources that say otherwise, I'd love to see them.
- brookst 4y ago
  Product placement is everywhere. Next time you watch a movie or show, look for the clothing brands, computer brands, car brands, wine brands, etc. everywhere. And think about sponsorships. From soccer to nascar, sports is covered with branding.
  - Spivak 4y ago
    That's not subliminal, you're describing sponsorships (i.e. manufactured social proof).
    - brookst 4y ago
      "Subliminal" and "sponsorship" are totally orthogonal. One refers to the presentation, the other the business arrangement.
- 4y ago

cmrdporcupine 4y ago

My guess is your first point is exactly why Google hasn't done this yet. Their 'knowledge boxes' are already crossing a line that in general they felt nervous about crossing historically, but they don't go very far. Google on the whole historically did not want to alienate publishers (and the advertisers that hang out on publisher content) and has avoided being in the content production business for this reason.

wongarsu 4y ago

If language models take over text content, content creators will flee even quicker into creating video content. There's already a trend where younger people tend to prefer video for being more "genuine", and now it might become a sign of "human made" for a couple years. Also easier to monetize, and easier to build parasocial relationships, so all around a plus for creators. Too bad I prefer text.

ikt 4y ago
could it somehow get access to the subtitles and then use them to answer queries? also i hope this comes to ecosia, would like to experiment and try it at least
- jadbox 4y ago
  Exactly, and many bots exist today to mine user videos for the automated subtitle information. In other words, there's no escaping GPT from learning from any kind of medium.
- throwaway09223 4y ago
  > " could it somehow get access to the subtitles and then use them to answer queries?" It's not even necessary - computers are already excellent at understanding spoken words. Have you tried automatic captioning recently? Half the inputs to my phone are already voice, not text. Video is a harder problem, but it's not too far behind.
belltaco 4y ago
I prefer text too but I feel like that's mostly because the videos are not information dense on purpose. They expand to whatever the youtube algorithm prefers at the time, which is about 10 minutes now. Ironically, tiktoks are more information dense but the search is completely useless.
- bronco21016 4y ago
  I’m finding more and more that the information density isn’t there because the video content is actually just an infomercial for a “course”.

WhiteNoiz3 4y ago

IMO this is the big problem with the internet as it exists today - there is no incentive for producing accurate, unbiased information and non-sensationalist opinions. My greatest hope for the future is that somehow we can incentivize people to produce "good" information for AI based assistants and move away from the rage/shock based advertising model that most of the internet currently uses. Personally I would rather pay a few cents for a query that produces valuable results and doesn't put me in a bad mood than pay with my time and attention like we do today. AI systems will absolutely need to be able to identify the training sources with every result (even if it is coming from several sources) and those sources should be compensated. IMO that's the only fair model for both image and text generation that is based on authors and artists work.

andsoitis 4y ago
> problem with the internet as it exists today - there is no incentive for producing accurate, unbiased information and non-sensationalist opinions. I think this problem is orthogonal to the internet as medium, though I’ll concede that it has proven to be the biggest amplifier of this dynamic. Correct (or correct as far as humans know, or most likely correct, etc.) costs money to create. False or completely made up information costs nothing, plus has the potential upside of sensationalism, thus further increasing its ROI. Agree with your point about developing more incentives for correct information and penalties for false.
int_19h 4y ago
It's not just that there's no incentive for that, but there's a very strong incentive to do the exact opposite: https://www.youtube.com/watch?v=rE3j_RHkqJc https://www.youtube.com/watch?v=rE3j_RHkqJc

dalbasal 4y ago

These questions are constant. I do think you bring up relevant issues, but they aren't quite showstoppers. Websites allow SE crawlers because (a) whatever traffic they get is better than not traffic (b) because allowing crawlers is default and doesn't cost anything and (c) google/bing don't negotiate. They are one, sites are many. This has already played out in news. News outlets wanted Google to pay for content. Google (initially) responded by allowing them to opt out of Google. Over the years, they have negotiated a little bit. Courts, in some places, forced Google to negotiate... It's news and politicians care about news specifically. Overall though, there have not been meaningful moments where people got pissed off with Google and blocked crawlers. Not newspapers and not anyone else. Site owners being mad doesn't affect google or Bing. What does matter to search engines is walled gardens. Facebook pioneered this, and this does matter to Google. There is, in a lot of cases, a lot less content to index and serve users. All those old forums, for example. These are search problems, and GPT-based search will inherit them. ChatGPT will have the same problem recommending the best air fryer as normal search does. GPT is a different way of presenting information... it's not presenting different information. RE: Lawsuits. Again, history. Youtube, for example, started off with rampant copyright infringement. But, legal systems were primitive. Lawyers and legislatures didn't know what to do. Claimants were extremely dispersed, and would have had to pioneer case law. Ultimately, copyright took >10 years to really apply online and by that point youtube and other social media was entrenched. The law lags. In practice, early movers are free to operate flawlessly and they get to shut the door after them. Now that Google is firmly entrenched, copyright law serves as one of their trenches.

jredwards 4y ago
Incidentally, law seems like an incredibly powerful potential application for ChatGPT.

VLM 4y ago

> Where is that information going to come from, exactly? Manufacturers, with quality ranging from excellent to trash. Consider trying to buy a 1K resistor at Digikey using their parametric search. Possible, but tedious and time consuming because you need a lot of domain knowledge to know what you want, and the technological range of "things with 1K of resistance" is extremely vast. At least its possible because the mfgrs are honest when Digikey imports their data. Consider the opposite, consumer goods. 500 watt PC power supplies with random marketing number stickers on the same chassis ranging from 500 to 1200 watts. Consumer level air compressors and consumer level vacuum cleaners than plug into household wall outlets claiming "8 horsepower" or whatever insane marketing nonsense. Clothes with vanity sizing so a "medium" tag fits like a real world XXL. Every processed food in a store with a "keto" label is high carb sugar added garbage, much like happened with "organic" label in the old days (the employees at the farm, distributor, warehouse, and/or retail store level take the same produce out of one bin and put it in two places with different prices) I think it will help when purchasing technical engineering type products but be an epic fail at inherently misleading consumer goods.

Dwolb 4y ago
Maybe manufacturers could upload their design docs and ChatGPT could learn exactly what the object does and what its performance parameters are.
Xelynega 4y ago
If you're trying to search for a specific resistor without the prerequisite domain knowledge, how will you be able to vet whether or not the answer given by a language model meets your needs? Imagining that language models like gpt will ever be able to index up-to-date information is literally trying to apply the concept of "artificial intelligence" to a probabilistic language model. It's incompatible with what it's actually doing.

ctoth 4y ago

Oh gosh, maybe we'll actually have to pay for things, and we'll find that the market for the fifth random blog trying to make money off of free information using ads doesn't really exist. What a terrible world this will obviously be. No. The weird thing is this idea that because you put ads on your site, you deserve money. Your ads are making the Internet worse. You probably don't realize this, because you most-likely use an ad blocker, which means you want people too dumb to use ad blockers to subsidize the web that you can use for free, but the current web is working well for approximately no one. Would I pay $5 a month for StackOverflow if it didn't show up for everything I Google? Most likely. Would this be a better world? almost certainly. We tried the thing with ads. It sucks. I welcome our new AI search overlords.

skinnymuch 4y ago
Why would you want power centralized? Big corporations are never your friend.
- int_19h 4y ago
  Power is also centralized when most supposedly independent actors buy ads from the same large advertisers, and utterly depend on their income from those ads to do whatever they're doing.

sieabahlpark 4y ago

[dead]

schaefer 4y ago

How I’d like to weather this storm: 1) Everyone cryptographically signs their work for identity confirmation. 2) there exists a blockchain whose sole purpose is to allow content creators to establish copyright date on a digital piece of work. 3) a public that uses the two items above what evaluating the reputation of an artist.

p0pcult 4y ago
This seems to make a lot of sense. The artists themselves also have an incentive to be blockchain validators/miners, thereby reducing the need for token payout, and the subsequent speculation that comes with tokenization (I think).
codethief 4y ago
How does that prevent anyone from using ChatGPT to generate new (supposedly human-written) content?
- p0pcult 4y ago
  It doesn't; however, the signature in your hypothetical doesn't correspond to a known/trusted author.
marwis 4y ago
You don't need blockchain for cryptographic timestamp
- schaefer 4y ago
  I've got some reading to do[1]. Thank you for the head's up. [1] https://en.wikipedia.org/wiki/Trusted_timestamping https://en.wikipedia.org/wiki/Trusted_timestamping

rmbyrro 4y ago

Their plan is to use Neuralink to pull all the information from people's brains.

_huayra_ 4y ago

> Barely anyone is making money writing reviews today, most sites are farmed content. I'm sure ChatGPT will be able to write a bunch of terrible SEO prose that precedes the actual air fryer review (or worse, recipe) about how the author's grandma had an air fryer when she was young and remembered the great times with her grandma (etc), for roughly 95% of the text! In all seriousness, being able to swerve all that terrible SEO content on reviews will always be welcome!

waynesonfire 4y ago

> Why would I ever let bing crawl my site if they aren't going to send any visitors to me? You tell me! It's your site. If you want money maybe you should charge for your content? And honestly, the web that Google presents is just so terrible that I don't want to visit your site, unfortunately. And, maybe it's a price worth paying.

throwaway4aday 4y ago

A language model that provides answers with sources (which could be found using a traditional search engine that searches the corpus that the language model is trained on) would be very useful and would also allow it to link directly to the source material. The trouble would be in finding the exact sources since the output of the language model is unlikely to be verbatim but current search engines can deal with imprecise queries fairly well so it's not an intractable problem. A very well curated data set would help this immensely. I'd be super interested in a language model that was able to synthesize knowledge drawn from a large corpus of books and then cite relevant sections from various titles.

bioemerl 4y ago

> Why would I ever let bing crawl my site if they aren't going to send any visitors to me? How can you refuse? The only way I know would be to require an account, but even then they could bypass it.

rafale 4y ago
Major search engines honor robots.txt
- bioemerl 4y ago
  If it becomes standard to have such a file and it effects their bottom line, could they disregard it?
  - Xelynega 4y ago
    If the file that were previously honored as consent to use the copyright material is subsequently ignored, wouldn't the content creators take the indexers to court for copyright infringement?

danielrhodes 4y ago

I don't see why this would be a copyright violation anymore than somebody learning something from multiple sources and reformulating what they learned into an answer to a question. As long as it isn't explicitly reciting its training data, there shouldn't be an issue of copyright.

sebzim4500 4y ago

>- Why would I ever let bing crawl my site if they aren't going to send any visitors to me? They will send at least some visitors, which is better than the zero visitors you will get from bing if you block it. >- fair use of snippets has relied on them being brief and linking to the source. Lawsuits will be immediate. Yes and microsoft has lawyers, who have presumably determined that the cost of fighting these frivolous lawsuits is not overwhelming.

6gvONxR4sf7o 4y ago

Put SEO into the picture and things get hairier. Incredibly realistic spam is about to go through the roof, so search engines will have an insanely harder time distinguishing between useful content and spam. Making money from search traffic to your (presumably useful) site is going to get harder in a bunch of ways, due to generative models.