The Coming Enshittification of AI

The article assumes that AI will get shitty the same way search, social media, etc. did. I don't think this is the case at all. It's pretty clear that LLMs will rapidly discover entirely new and innovative ways to be shitty.

It's unclear to me how this will play out because LLMs don't have the same network / platform effects as the other examples (Uber / Facebook), nor is there one dominant LLM that is overwhelmingly better than the competition for consumers (Google). There's overwhelming competition from the open source cheap models especially for the lower-mid intelligence use cases

I have never been forced to watch an add on the Facebook app nor the Instagram app but the moment that happens to me I will uninstall both immediately. At this point Facebook is lamer/worse than MySpace ever was.

Microsoft has already started. Copilot is now sitting behind a forced full browser window login popup.

And I do really mean forced popup, because all default functionality - which is still available to a anonymous user - is available right behind said popup, which I can clearly see due to the few milliseconds it takes before their shitty React code triggers the visibility of said popup on my secondary laptop.

This has always and will always be my fundamental concern regarding AI. It will inevitably be enshittified, and its enshittification will be both subtle and opaque, in ways the average users will be ill equipped to identify or avoid.

The ultimate problem will be, unless users are running their own local LLMs (which will of course perpetually remain a tiny, insignificant fraction of all AI users) the AI isn't going to be working in the interests of the user, but the rather the interest of the corporation that built it, and the advertisers that pay said corporation.

At least with social or search there are ad standards that the platforms all broadly follow that identify when something is an ad, where as the AI companies will almost certainly either ignore ad standards guidelines or find some loophole they feel justifies not disclosing to users when the answers their AI provides are effectively ads in disguise.

With search, if you search for say "Cheapest car insurance" you at least have a fighting chance of successfully determining which links are ads, which are spam and which are potentially useful content, but with AI, its just going to provide a single supposedly authoritative answer, that the user will think is the best answer to their question, but is in fact whichever answer an advertiser has paid the AI company to have the AI supply.

Local llm is so important to avoid ads, because the future of AI is tons of ads

This article describes well how advertising and such degrades the quality of the product, but it isn’t really enshittification unless the platform is also turning the screw on its suppliers. This might be the case for services downstream of frontier model providers, but OpenAI and friends aren’t turning the screw on NVidia or utilities.

If anything them trying to do the customer-side enshittification before securing both ends of their market is just a sign that they’re troubled and will likely not ever reach that point.

I dispute that. Enshitification happens when the user does not pay for the product. The provider of the service has to find alternative ways of monetization. As a user, this feels shitty, but you get what you pay for. With AI, if you pay for usage, you can demand that the service provided not be shitty. You can even litigate, but the most important leverage is the threat to take your business elsewhere. Enough people do that and the provided service stops being shitty in a hurry.

> I’m in constant awe at what LLMs have brought us in terms of speed and how they act as a force multiplier when used cleverly, but previous experiences with disruptive technologies that shaped society (like the Internet, search engines and social networks) show us that these will inevitably be turned into ad-selling machines, enshittified like everything before them.

IMO, if that's the case, then it's yet another example of "We can't get anyone to pay for this stuff"

So, it sediments and is used to sell "real" things, ones that people actually pay money for, like phones. Thinking about this, phones, computers, toasters, thermostats, you name it, becomes not a "bicycle for the mind" but a "chauffeur for the mind" ... you don't have to think; just go along, and "things" just become gateways to some replacement mind out there.

Being such, their value decreases to roughly zero, and (oops) that external mind out there makes all your decisions for you, including what to buy that's of real value, or political decisions.

It all starts with ceding decisions to some outside source rather than using an outside source for information with which to make decisions. Let's face it, trusting humans to help make decisions has always been risky business. Trust is earned, and you can at least look a human in the face, for what that's worth.

Now, humans compete for your dollar, and depending on your politics/philosophy, that either gets you better products and services through competition or is a giant waste that depletes resources through duplication and leads to exaggerated claims (lies at times). Now, extend that to AI "agents" that left unchecked, can steer you wrong. So you set up a competing agent, and the world fills up with "AI wars".

"AI, find a way to swindle Pop out of his home and belongings"

"AI, find a way to protect Pop from swindlers"

This Pop just needs friends he can trust. We are problem-solvers by nature. It keeps us sharp. If we could just cooperate more, life would be better. Don't forget, the forces of racism, nationalism, religious enmity are propaganda, with no basis in our fundamental nature, and they create "stories" and color facts. Oh, that's advertising. In French, avertissement means warning.

Cooperation implies and requires trust. Competition usually involves coercion of some sort.

First, they poison information (Google, you bastards) and then "steer" you wrong. I use that word because, much as I could use a fully self-driving auto, I also believe that if I ask it to go to the place I like for pizzas, it will lock the doors and take over steering, and head out to the one it is paid to.

With trust, you don't need "control"

When will we learn that?