5 ms·

I've been having a lot of success using o3 to run searches. Its really nice to be able to parse through tons of search results and just get the relevant info (p

I've been having a lot of success using o3 to run searches. Its really nice to be able to parse through tons of search results and just get the relevant info (probably what the search engine should have been doing in the first place, but I digress). I really don't want to have to give this up, but I imagine soon enough this too will become enshittified. I mean, its already happening: https://openai.com/chatgpt/search-product-discovery/ https://openai.com/chatgpt/search-product-discovery/ Whats the long term solution here? Open Web UI with deepseek + tavily? Would it be profitable long term to have a "neutral" search engine, or will it be cost prohibitive moving forward?

swores 1y ago
> I imagine soon enough this too will become enshittified. I mean, its already happening: https://openai.com/chatgpt/search-product-discovery/ https://openai.com/chatgpt/search-product-discovery/ For now, at least, OpenAI claim that those product suggestions (almost tempted to leave in my typo / phone's autocorrect of "subversions") are not ads, and that it's purely a feature designed to be useful for ChatGPT users. Although this from the FAQ is a bit strange, and I do wonder if there's any business relationship between OpenAI and the "third party providers" that happens to involve money passing from the latter to OpenAI in commercial deals that are definitely not ad purchases... > How Merchants Are Selected > When a user clicks on a product, we may show a list of merchants offering it. This list is generated based on merchant and product metadata we receive from third-party providers. Currently, the order in which we display merchants is predominantly determined by these providers. We do not re-rank merchants based on factors such as price, shipping, or return policies. We expect this to evolve as we continue to improve the shopping experience. > To that end, we’re exploring ways for merchants to provide us their product feeds directly, which will help ensure more accurate and current listings. If you're interested in participating, complete the interest form here, and we’ll notify you once submissions open. ( https://help.openai.com/en/articles/11128490-improved-shopping-results-from-chatgpt-search https://help.openai.com/en/articles/11128490-improved-shoppi... )
__MatrixMan__ 1y ago
I think the long term solution is P2P search where the client is configured to know who you trust. So if you search for waffles, and you trust somebody who trusts somebody who published a waffle recipe, that'll probably be one of your first results. If you get an untrustworthy result, revoke trust in whoever caused you to see it. It's not exactly on the horizon but I think it's possible to build a web which rewards being trustworthy, rather than one that rewards attention mongering.
- headcanon 1y ago
  If I created a curated list of known "good" websites that I find "trustworthy" and build a search index just with the info there, would that satisfy this? What else would need to be built?
  - __MatrixMan__ 1y ago
    I think it makes more sense if it's people that we find trustworthy, rather than websites. So there would need to be some place the client could look to find all of that person's content, regardless of which site it was posted on (something along the lines of https://solidproject.org/ https://solidproject.org/). Or if not a copy of the content, then at least a hash and a signature. So for instance if you're browsing reviews on amazon you could filter to reviews in your trust network, and your browser would match the content to what your peers have published and verify that yes indeed, they wrote that review. It's gotta be hosted elsewhere so Amazon can't tamper with it. Also, in order for it to have enough reach I think it would have to be recursive, so it's not just the stuff your explicitly trusted peers have posted/endorsed, but all the stuff from their peers, and their peers, and so-on. Sometimes this implicit FoaF chain could be quite long without being a problem, but if you start running across harmful content, you could explicitly prune the tree in those places, and your explicit distrust would persist so that those parties couldn't reach you thereafter.
satisfice 1y ago
You say you’ve been having a lot of success… how can you possibly know that? Whenever I get any summary or diatribe or lecture out of a chatbot all I know is that I have a major fact checking challenge. And I don’t have time for it. I cannot believe you are doing all that fact-checking.
- istjohn 1y ago
  Ironic username
  - satisfice 1y ago
    Is it?
- headcanon 1y ago
  It includes website references for most of these searches so its easily verifiable. Here's an example: https://chatgpt.com/share/6839b2a0-d4f4-8000-9224-f4065898029f https://chatgpt.com/share/6839b2a0-d4f4-8000-9224-f406589802... I was traveling in Tokyo recently and took a picture of a van that was hosting what looked like a political rally in Akihabara with hand painted slogans on the outside. It wrote some python code to analyze the image segment by segment and eventually came up with the translations. Then it was able to find me the website for the political party, which had an entry for the rally that was held that day. I don't speak Japanese so its possible some of the translations were not accurate but it looked like it generally lined up and it ultimately got me what I wanted eventually. I was there a year ago as well and tried doing similar translations and it had a very hard time with the hand painted kanji. Its really come a long way since then. I also used it to find some obscure anime events the same day, most of which are only announced in Japanese on obscure websites. Being a non-speaker and not familiar with the websites it would have been a huge pain to google.
  - satisfice 1y ago
    If you are using AI to do things that you can immediately verify, that’s cool. A great many things AI can supposedly do are not easily verifiable. A client of mine used it to “summarize” a series of lectures I did. It got a lot right and crucial things wrong that required me to warn them against trusting any of the summaries. They were fatally contaminated in a way that required damage control.

swores 1y ago

> I imagine soon enough this too will become enshittified. I mean, its already happening: https://openai.com/chatgpt/search-product-discovery/ https://openai.com/chatgpt/search-product-discovery/ For now, at least, OpenAI claim that those product suggestions (almost tempted to leave in my typo / phone's autocorrect of "subversions") are not ads, and that it's purely a feature designed to be useful for ChatGPT users. Although this from the FAQ is a bit strange, and I do wonder if there's any business relationship between OpenAI and the "third party providers" that happens to involve money passing from the latter to OpenAI in commercial deals that are definitely not ad purchases... > How Merchants Are Selected > When a user clicks on a product, we may show a list of merchants offering it. This list is generated based on merchant and product metadata we receive from third-party providers. Currently, the order in which we display merchants is predominantly determined by these providers. We do not re-rank merchants based on factors such as price, shipping, or return policies. We expect this to evolve as we continue to improve the shopping experience. > To that end, we’re exploring ways for merchants to provide us their product feeds directly, which will help ensure more accurate and current listings. If you're interested in participating, complete the interest form here, and we’ll notify you once submissions open. ( https://help.openai.com/en/articles/11128490-improved-shopping-results-from-chatgpt-search https://help.openai.com/en/articles/11128490-improved-shoppi... )

__MatrixMan__ 1y ago

I think the long term solution is P2P search where the client is configured to know who you trust. So if you search for waffles, and you trust somebody who trusts somebody who published a waffle recipe, that'll probably be one of your first results. If you get an untrustworthy result, revoke trust in whoever caused you to see it. It's not exactly on the horizon but I think it's possible to build a web which rewards being trustworthy, rather than one that rewards attention mongering.

headcanon 1y ago
If I created a curated list of known "good" websites that I find "trustworthy" and build a search index just with the info there, would that satisfy this? What else would need to be built?
- __MatrixMan__ 1y ago
  I think it makes more sense if it's people that we find trustworthy, rather than websites. So there would need to be some place the client could look to find all of that person's content, regardless of which site it was posted on (something along the lines of https://solidproject.org/ https://solidproject.org/). Or if not a copy of the content, then at least a hash and a signature. So for instance if you're browsing reviews on amazon you could filter to reviews in your trust network, and your browser would match the content to what your peers have published and verify that yes indeed, they wrote that review. It's gotta be hosted elsewhere so Amazon can't tamper with it. Also, in order for it to have enough reach I think it would have to be recursive, so it's not just the stuff your explicitly trusted peers have posted/endorsed, but all the stuff from their peers, and their peers, and so-on. Sometimes this implicit FoaF chain could be quite long without being a problem, but if you start running across harmful content, you could explicitly prune the tree in those places, and your explicit distrust would persist so that those parties couldn't reach you thereafter.

satisfice 1y ago

You say you’ve been having a lot of success… how can you possibly know that? Whenever I get any summary or diatribe or lecture out of a chatbot all I know is that I have a major fact checking challenge. And I don’t have time for it. I cannot believe you are doing all that fact-checking.

istjohn 1y ago
Ironic username
- satisfice 1y ago
  Is it?
headcanon 1y ago
It includes website references for most of these searches so its easily verifiable. Here's an example: https://chatgpt.com/share/6839b2a0-d4f4-8000-9224-f4065898029f https://chatgpt.com/share/6839b2a0-d4f4-8000-9224-f406589802... I was traveling in Tokyo recently and took a picture of a van that was hosting what looked like a political rally in Akihabara with hand painted slogans on the outside. It wrote some python code to analyze the image segment by segment and eventually came up with the translations. Then it was able to find me the website for the political party, which had an entry for the rally that was held that day. I don't speak Japanese so its possible some of the translations were not accurate but it looked like it generally lined up and it ultimately got me what I wanted eventually. I was there a year ago as well and tried doing similar translations and it had a very hard time with the hand painted kanji. Its really come a long way since then. I also used it to find some obscure anime events the same day, most of which are only announced in Japanese on obscure websites. Being a non-speaker and not familiar with the websites it would have been a huge pain to google.
- satisfice 1y ago
  If you are using AI to do things that you can immediately verify, that’s cool. A great many things AI can supposedly do are not easily verifiable. A client of mine used it to “summarize” a series of lectures I did. It got a lot right and crucial things wrong that required me to warn them against trusting any of the summaries. They were fatally contaminated in a way that required damage control.