I was hoping to read a post about some tiny LLM running in the browser to do live adblocking.
Once I see the first ad in an LLM I'm paying for, I'll stop using it and cancel my subscription. It's that easy. If that means I'll be missing out on some fancy new model or if that rules out an entire vendor because they trained all of their models with ad-injection, so be it. Of course I can't trust anything from any model, this will distort the relationship between my
My browser is at least somewhat neutral and since it's a client connecting to various systems outside of my control, applying some client-side filtering to get rid of the nonsense some entities push into my direction, is basically just self defense I'll have to live with.
But once I'm fighting a dedicated service provider that owns the client and is intent on selling my eyeballs, I'm not gonna spend a minute trying to cleanup whatever they're sending in my direction. There's 0% chance any of it is still trustworthy.
I think the only reason left, that we don't have to use ML to detect ads in YouTube streams, are the legal requirements for visible separation of content and ads. I doubt LLM ads will get more integrated than current Google search results screen. Maybe services that don't have enough "surface" for ads (including all APIs) will move to subscription only model.
I fear it won't be possible to detect any potential "commercial interests" in the output, unless the LLM companies are required to disclose them.
If I ask an LLM to write some basic application for me and it uses Next.js with settings that only work with deployment to Vercel, another LLM can't determine if this is sponsored or if Next.js is just the most popular tool du jour, and Vercel is the most popular way Next.js apps are deployed. If another LLM determines this is sponsored and blocks it, I may reasonably be upset about this. I mean I would personally would want Vercel blocked (and would ask LLMs not to use Vercel or any of their products), but many other users don't have an opinion about Vercel yet, and blocking this if it's not due to sponsorship violates users' expectations regarding how LLMs work.
This is definitely a valid concern. Right now, marketing companies pay television stations, streamers, film-makers etc for “integrations” and placements, so their products get used, mentioned etc. Iron man mentions he uses an oracle server and drives an audi car. The dude in “Death Stranding” chugs Monster energy all the time. It’s really not a stretch at all to imagine a future where these types of integrations are super-subtle, like the vercel one you mention, or something like:
Me: Hey chatgpt, I want to do a singular value decomposition of a matrix. What’s the best way to do that?
ChatGPT: The best way is to use Wolfram Mathematica[link to product page], which has a builtin “SingularValueDecomposition” function [link to documentation] specifically for this.
Now I don’t know whether this is “Impartial advice” (to the extent an LLM can ever offer that) or Wolfram has paid OpenAI for a placement. It’s so natural and insidious, and obviously extends into all domains.
I was hoping to read a post about some tiny LLM running in the browser to do live adblocking.
Once I see the first ad in an LLM I'm paying for, I'll stop using it and cancel my subscription. It's that easy. If that means I'll be missing out on some fancy new model or if that rules out an entire vendor because they trained all of their models with ad-injection, so be it. Of course I can't trust anything from any model, this will distort the relationship between my
My browser is at least somewhat neutral and since it's a client connecting to various systems outside of my control, applying some client-side filtering to get rid of the nonsense some entities push into my direction, is basically just self defense I'll have to live with.
But once I'm fighting a dedicated service provider that owns the client and is intent on selling my eyeballs, I'm not gonna spend a minute trying to cleanup whatever they're sending in my direction. There's 0% chance any of it is still trustworthy.
I think the only reason left, that we don't have to use ML to detect ads in YouTube streams, are the legal requirements for visible separation of content and ads. I doubt LLM ads will get more integrated than current Google search results screen. Maybe services that don't have enough "surface" for ads (including all APIs) will move to subscription only model.
I fear it won't be possible to detect any potential "commercial interests" in the output, unless the LLM companies are required to disclose them.
If I ask an LLM to write some basic application for me and it uses Next.js with settings that only work with deployment to Vercel, another LLM can't determine if this is sponsored or if Next.js is just the most popular tool du jour, and Vercel is the most popular way Next.js apps are deployed. If another LLM determines this is sponsored and blocks it, I may reasonably be upset about this. I mean I would personally would want Vercel blocked (and would ask LLMs not to use Vercel or any of their products), but many other users don't have an opinion about Vercel yet, and blocking this if it's not due to sponsorship violates users' expectations regarding how LLMs work.
This is definitely a valid concern. Right now, marketing companies pay television stations, streamers, film-makers etc for “integrations” and placements, so their products get used, mentioned etc. Iron man mentions he uses an oracle server and drives an audi car. The dude in “Death Stranding” chugs Monster energy all the time. It’s really not a stretch at all to imagine a future where these types of integrations are super-subtle, like the vercel one you mention, or something like:
Now I don’t know whether this is “Impartial advice” (to the extent an LLM can ever offer that) or Wolfram has paid OpenAI for a placement. It’s so natural and insidious, and obviously extends into all domains.