5 ms·

AI is following the drug dealer model. “The first dose is free!” Given the cost incurred, lots of dark patterns will be coming for sure.

littlecranky67 1y ago
I fail to see how that will work out. Just I have an adblocker now, I could have a very simple local llm in my browser that modifies the search-AIs answer and strips obvious ads.
- svachalek 1y ago
  They won't be obvious. They'll be highly customized brain worms influencing your votes and purchases to the highest bidder.
  - A4ET8a8uTh0_v2 1y ago
    Yep. Right now, even with cookies inferences about individual humans are minimal, but exposing your whole patterns in speech make you a ripe target for manipulation at a scale that some may not fully understand yet. 4o is very adept at cold reading and it is genuinely fascinating to read those from that perspective alone. Combine it with style evaluation and a form of rudimentary history analysis, and you end up with actual dossier on everyone using that service. Right now, we are lucky, because it is the least altered version of it ( and we all know how many filters public models have to go through ).
  - lopis 1y ago
    Which sounds very illegal in most places, as clearly identifying sponsored content is required. Let's see how that turns out.
    - _DeadFred_ 1y ago
      For hire drivers/having employees had very specific legal requirements in most areas. Let's see how that turned out. Oh yeah, the dystopian tech companies won and we the people got the benefit job/job rules being thrown out for the beauty that is independent contractor 'gig work'.
    - floatrock 1y ago
      > as clearly identifying sponsored content is required Citation needed? Once AI content becomes monetized with ads, it's not going to look like the ads/banners we're used to. If you're looking into the past, you don't understand the potential of AI. Noam Chomsky's manufactured consent is going to look quaint by comparison.
throwaway290 1y ago
Yep. Dark patterns you can see are not that dark by comparison, we will need another word for coming dark patterns disguised in llm responses
- lelanthran 1y ago
  > Yep. Dark patterns you can see are not that dark by comparison, we will need another word for coming dark patterns disguised in llm responses As someone else said, you can probably filter responses through a small purpose-built/trained LLM that strips away dark patterns. If you start getting mostly empty responses as a result, then there was no value prior to the stripping anyway.
  - throwaway290 1y ago
    If you can't tell when a big expensive llm is subliminally grooming you to like/dislike something or is selective with information then another (probably small and cheaper) llm somehow can? Arms race?
    - lelanthran 1y ago
      > If you can't tell when a big expensive llm is subliminally grooming you to like/dislike something or is selective with information I mean, I can tell when a page contains advertisements, but I still use an ad-blocker. The point was not to help me detect when a response is ad-heavy, but to stop me seeing those ads at all. > Arms race? Possibly. Like with ad-blockers, this race can't be won by the ad-pusher LLM if the user uses the ad-blocker LLM. The only reason ad-pusher websites still work is because users generally don't care enough to install the ad-blocker. In much the same way, the only reason LLM ad-pushers will work is if users don't bother with an LLM ad-blocker.
      - throwaway290 1y ago
        > I mean, I can tell when a page contains advertisements, but I still use an ad-blocker Yep, because why let people make money from their work right? You should just get content for free! > this race can't be won by the ad-pusher LLM if the user uses the ad-blocker LLM. As per my comment it literally can.
jdietrich 1y ago
It's a market where nobody has a particularly deep moat and most players are charging money for a service. Open weight models aren't too far behind proprietary models, particularly for mundane queries. The cost of inference is plummeting and it's already possible to run very good models at pennies per megatoken. I think it's unreasonably pessimistic to assume that dark patterns are an inevitability.
- simgt 1y ago
  For the sake of argument, none of the typical websites with the patterns described have a moat, and the cost of hosting them has plummeted a while ago. It's not inevitable but it's likely, and they will be darker if they are embedded in the models' output...
- azangru 1y ago
  > and most players are charging money for a service The aricle talks about AI overviews. As exemplified by the AI summary at the top of Google search results page. That thing is free.
  - svachalek 1y ago
    1. Create free and good product 2. Attract large user base 3. Sell user data and attention to advertisers 4. Extract maximal profit from sponsors 5. Earn billions from shit product
    - geerlingguy 1y ago
      Hey that's like a popular Search engine's search results page!
- ToucanLoucan 1y ago
  You do realize of course that every service that now employs all these dark patterns we're complaining about was already profitable and making good money, and that simply isn't good enough? Revenue has to increase quarter-to-quarter otherwise your stock will tank. It's not simply enough that a product "makes money" it must "make more money, every quarter, forever" which is why everything, not even limited to tech, but every product absolutely BLOWS. It's why every goddamn thing is a subscription now. It's why every fucking website on the internet wants an email and a password so they can have you activate an account, and sell a known active email to their ad partners.
bdelmas 1y ago
Well maybe not. Thanks that we have Gemini now to compete with ChatGPT. Competition may avoid dark patterns. But without competition yes definitely
- generic92034 1y ago
  Competition or not, dark patterns or not - sooner or later LLMs will need to earn money for their corporations.
  - moontear 1y ago
    But they do? Paid subscriptions for Gemini, ChatGPT and Copilot are a thing. If Google throws in a free AI summary in their search it only helps promoting Gemini in the long run.
    - ileonichwiesz 1y ago
      Look up the numbers. OpenAI actually loses money on every paid subscription, and they’re burning through billions of dollars every year. Even if you convince a fraction of the users to pay for it, it’s still not a sustainable model.
      - nicbou 1y ago
        Even if they were profitable, the investors would feel that it's not profitable enough. They won't stop at breaking even.
        generic92034 1y ago
        And even if it was the highest profit branch of the company, they still would see a need to do anything possible to further increase profits. That is often where enshittification sets in. This currently is the sweet phase where growing and thus gaining attention and customers as well as locking in new established processes is dominant. Unless the technical AI development stays as fast as in the beginning, this is bound to change.
nicbou 1y ago
AI is built by the same companies that built the last generation of hostile technology, and they're currently offering it at a loss. Once they have encrusted themselves in our everyday lives and killed the independent web for good, you can bet they will recoup on their investment.
- A4ET8a8uTh0_v2 1y ago
  That indeed is likely to come, but having experienced user hostile technology, the appropriate response is to prepare. Some trends suggest this is already happening ( though that only appears to be a part HN crowd so far ): moving more and more behind a local network. I know I am personally exploring local llm integration for some workflows to avoid the obvious low hanging fruit most providers will likely go for. But yes, the web in its current form might perish.
  - _DeadFred_ 1y ago
    Would be cool if local libraries got together and figured out how to allow access to community LLMs. That fits more with my idea of the future and AIs than having the now dystopian tech companies running/defining it all.
- pcdoodle 1y ago
  Is there another edge to this sword? Can we fight back with LLMs that ignore sources with all the tracking / SEO and other garbage? I'd love to tell my local LLM that "I hate pintrest" for instance and it just goes "okay, pintrest shields are up".
  - fxtentacle 1y ago
    Kagi allows you to block Pinterest
    - a_vanderbilt 1y ago
      Kagi got me to sign up as an early adopter because it let me banish pinterest forever.
deadbabe 1y ago
To combat this, maybe we can cache AI responses for common prompts somehow and make some kind of website where people could search for keywords and find responses that might be related to what they want, so they don’t have to spend tokens on an AI. Could be free.
- chasd00 1y ago
  I would be curious to see what would happen if you could write every query/response from an LLM to an HTML file and then serve that directory of files back to google with a simple webserver for indexing.
  - deadbabe 1y ago
    I think the future will be: 1. Someone prompts 2. Server searches for equivalent prompts, if something similar was asked before, return that response from cache. 3. If prompt is unique enough, return response from LLM and cache new response. 4. If user decides response isn’t specific enough, ask LLM and cache.

littlecranky67 1y ago

I fail to see how that will work out. Just I have an adblocker now, I could have a very simple local llm in my browser that modifies the search-AIs answer and strips obvious ads.

svachalek 1y ago
They won't be obvious. They'll be highly customized brain worms influencing your votes and purchases to the highest bidder.
- A4ET8a8uTh0_v2 1y ago
  Yep. Right now, even with cookies inferences about individual humans are minimal, but exposing your whole patterns in speech make you a ripe target for manipulation at a scale that some may not fully understand yet. 4o is very adept at cold reading and it is genuinely fascinating to read those from that perspective alone. Combine it with style evaluation and a form of rudimentary history analysis, and you end up with actual dossier on everyone using that service. Right now, we are lucky, because it is the least altered version of it ( and we all know how many filters public models have to go through ).
- lopis 1y ago
  Which sounds very illegal in most places, as clearly identifying sponsored content is required. Let's see how that turns out.
  - _DeadFred_ 1y ago
    For hire drivers/having employees had very specific legal requirements in most areas. Let's see how that turned out. Oh yeah, the dystopian tech companies won and we the people got the benefit job/job rules being thrown out for the beauty that is independent contractor 'gig work'.
  - floatrock 1y ago
    > as clearly identifying sponsored content is required Citation needed? Once AI content becomes monetized with ads, it's not going to look like the ads/banners we're used to. If you're looking into the past, you don't understand the potential of AI. Noam Chomsky's manufactured consent is going to look quaint by comparison.

throwaway290 1y ago

Yep. Dark patterns you can see are not that dark by comparison, we will need another word for coming dark patterns disguised in llm responses

lelanthran 1y ago
> Yep. Dark patterns you can see are not that dark by comparison, we will need another word for coming dark patterns disguised in llm responses As someone else said, you can probably filter responses through a small purpose-built/trained LLM that strips away dark patterns. If you start getting mostly empty responses as a result, then there was no value prior to the stripping anyway.
- throwaway290 1y ago
  If you can't tell when a big expensive llm is subliminally grooming you to like/dislike something or is selective with information then another (probably small and cheaper) llm somehow can? Arms race?
  - lelanthran 1y ago
    > If you can't tell when a big expensive llm is subliminally grooming you to like/dislike something or is selective with information I mean, I can tell when a page contains advertisements, but I still use an ad-blocker. The point was not to help me detect when a response is ad-heavy, but to stop me seeing those ads at all. > Arms race? Possibly. Like with ad-blockers, this race can't be won by the ad-pusher LLM if the user uses the ad-blocker LLM. The only reason ad-pusher websites still work is because users generally don't care enough to install the ad-blocker. In much the same way, the only reason LLM ad-pushers will work is if users don't bother with an LLM ad-blocker.
    - throwaway290 1y ago
      > I mean, I can tell when a page contains advertisements, but I still use an ad-blocker Yep, because why let people make money from their work right? You should just get content for free! > this race can't be won by the ad-pusher LLM if the user uses the ad-blocker LLM. As per my comment it literally can.

jdietrich 1y ago

It's a market where nobody has a particularly deep moat and most players are charging money for a service. Open weight models aren't too far behind proprietary models, particularly for mundane queries. The cost of inference is plummeting and it's already possible to run very good models at pennies per megatoken. I think it's unreasonably pessimistic to assume that dark patterns are an inevitability.

simgt 1y ago
For the sake of argument, none of the typical websites with the patterns described have a moat, and the cost of hosting them has plummeted a while ago. It's not inevitable but it's likely, and they will be darker if they are embedded in the models' output...
azangru 1y ago
> and most players are charging money for a service The aricle talks about AI overviews. As exemplified by the AI summary at the top of Google search results page. That thing is free.
- svachalek 1y ago
  1. Create free and good product 2. Attract large user base 3. Sell user data and attention to advertisers 4. Extract maximal profit from sponsors 5. Earn billions from shit product
  - geerlingguy 1y ago
    Hey that's like a popular Search engine's search results page!
ToucanLoucan 1y ago
You do realize of course that every service that now employs all these dark patterns we're complaining about was already profitable and making good money, and that simply isn't good enough? Revenue has to increase quarter-to-quarter otherwise your stock will tank. It's not simply enough that a product "makes money" it must "make more money, every quarter, forever" which is why everything, not even limited to tech, but every product absolutely BLOWS. It's why every goddamn thing is a subscription now. It's why every fucking website on the internet wants an email and a password so they can have you activate an account, and sell a known active email to their ad partners.

bdelmas 1y ago

Well maybe not. Thanks that we have Gemini now to compete with ChatGPT. Competition may avoid dark patterns. But without competition yes definitely

generic92034 1y ago
Competition or not, dark patterns or not - sooner or later LLMs will need to earn money for their corporations.
- moontear 1y ago
  But they do? Paid subscriptions for Gemini, ChatGPT and Copilot are a thing. If Google throws in a free AI summary in their search it only helps promoting Gemini in the long run.
  - ileonichwiesz 1y ago
    Look up the numbers. OpenAI actually loses money on every paid subscription, and they’re burning through billions of dollars every year. Even if you convince a fraction of the users to pay for it, it’s still not a sustainable model.
    - nicbou 1y ago
      Even if they were profitable, the investors would feel that it's not profitable enough. They won't stop at breaking even.
      - generic92034 1y ago
        And even if it was the highest profit branch of the company, they still would see a need to do anything possible to further increase profits. That is often where enshittification sets in. This currently is the sweet phase where growing and thus gaining attention and customers as well as locking in new established processes is dominant. Unless the technical AI development stays as fast as in the beginning, this is bound to change.

nicbou 1y ago

AI is built by the same companies that built the last generation of hostile technology, and they're currently offering it at a loss. Once they have encrusted themselves in our everyday lives and killed the independent web for good, you can bet they will recoup on their investment.

A4ET8a8uTh0_v2 1y ago
That indeed is likely to come, but having experienced user hostile technology, the appropriate response is to prepare. Some trends suggest this is already happening ( though that only appears to be a part HN crowd so far ): moving more and more behind a local network. I know I am personally exploring local llm integration for some workflows to avoid the obvious low hanging fruit most providers will likely go for. But yes, the web in its current form might perish.
- _DeadFred_ 1y ago
  Would be cool if local libraries got together and figured out how to allow access to community LLMs. That fits more with my idea of the future and AIs than having the now dystopian tech companies running/defining it all.
pcdoodle 1y ago
Is there another edge to this sword? Can we fight back with LLMs that ignore sources with all the tracking / SEO and other garbage? I'd love to tell my local LLM that "I hate pintrest" for instance and it just goes "okay, pintrest shields are up".
- fxtentacle 1y ago
  Kagi allows you to block Pinterest
  - a_vanderbilt 1y ago
    Kagi got me to sign up as an early adopter because it let me banish pinterest forever.

deadbabe 1y ago

To combat this, maybe we can cache AI responses for common prompts somehow and make some kind of website where people could search for keywords and find responses that might be related to what they want, so they don’t have to spend tokens on an AI. Could be free.

chasd00 1y ago
I would be curious to see what would happen if you could write every query/response from an LLM to an HTML file and then serve that directory of files back to google with a simple webserver for indexing.
- deadbabe 1y ago
  I think the future will be: 1. Someone prompts 2. Server searches for equivalent prompts, if something similar was asked before, return that response from cache. 3. If prompt is unique enough, return response from LLM and cache new response. 4. If user decides response isn’t specific enough, ask LLM and cache.