I need to preface this by saying that I’m in favor of LLMs and generative AI in general. I use them daily, and I think you should too.
My pessimism isn’t about the technology. It’s about the relationship we’re building with it.
The Google effect
Google has been the best search engine (it might not be anymore, but I’m not gonna argue about that). It built the most comprehensive index of information from across the world. It’s laughably simple to find almost anything you’re looking for.
I think many of us expected humanity to use this access to become more informed, smarter, and more skilled. You have practically unlimited access to any information you might ever want. People should be smarter and make better-informed decisions.
The exact opposite seems to have happened. A few people used that access to learn more. Most used it so they could be “dumber.” Why know things when I can just look them up when I need them?
This is compounded by the fact that the majority of people can’t actually Google for shit. They don’t know how to phrase what they are searching for, let alone use any kind of advanced search.
Media literacy is even worse. People struggle to recognize credible sources and critically evaluate the information they find.
I Googled it, so it must be true.
AI could make all of this worse.
How bad could it get?
Google happened. We are still here. It’s fine.
AI could be much, much worse. We don’t need AGI for that to happen.
The existing problems will compound, and AI introduces some new ones.
You still need to know how to ask
Google required you to describe what you were looking for. Using AI well requires even more. You don’t just ask for information. You command actions. You need to express your intent, define a goal, describe the requirements, and set limits.
If you are doing any agentic coding, you know what I’m talking about. You know people who prompt an agent with “doesn’t work” and expect a meaningful result.
I don’t need to know. AI can do it for me
This is no longer just about not knowing information. AI can take action, so you don’t need to know how to actually do things either.
You know less, and you understand even less about how things are done. We get even dumber and become increasingly dependent on systems we don’t understand.
How do you know it’s right?
If your ability to check its work is fundamentally compromised, how do you know when AI is hallucinating? Do you have a way to validate what it tells you? Are you still capable of validating it?
We already struggled to evaluate the results of a simple Google search. Now we have an LLM that can be prompted to agree with us, or given a system message that makes it behave in ways we aren’t privy to.
But it sounds so “human.” It wouldn’t lie.
How does a stone feel about it?
I heard George Hotz say that if you draw a smiley face on a stone, people will project emotions onto it.
We can already see this happening with AI. It’s the “weirdos” for now. But more of us will have AI friends, and trust them, than we would like to admit. People have always had a tendency to personify things. AI just gives those things a voice.
I didn’t delete production. AI did it!
You command it. You validate the result. You might even trust it. Who gets the credit, and who is responsible?
When you present work created with AI, did you really do it? Should you get the credit?
And when the AI fucks up? Damn that AI. It sometimes does strange things. It did it on its own.
Personally, I think the credit belongs to you as much as the responsibility does. You don’t get to claim one and reject the other.
Did you command AI to create something cool? Great! It wouldn’t have done anything on its own.
Did it delete the production database? That’s still on you. You created the environment and gave a probabilistic system enough information and access to produce a dangerous result.
You can already see people shifting blame to AI whenever something goes wrong. I think we’re about one or two years away from someone trying to put an LLM on trial.
Passion, identity, and reliance
I know I’m not the only one who doesn’t get the same feeling after making something with AI. There aren’t enough obstacles to trigger the same chemical reward in my brain. The more I use it, the more it feels like tying my shoelaces. I did something, sure, but it doesn’t feel like much of an accomplishment.
There is also a partial loss of identity, especially when part of who you are is built around what you can do. If AI can suddenly do much of it for you, what is left of that identity?
We will have to adapt. Building your identity entirely around a set of skills won’t hold up when those skills can be automated.
Reliance is the scariest part because it’s already happening, even if it isn’t visible yet. We are starting to offload not just parts of our work to AI, but parts of our lives.
This reminded me of a quote from Upgrade (watch it, it’s fun):
I don’t run this company anymore. I haven’t for years. Now I answer to someone. Someone much smarter than I ever could be.
Upgrade (2018)
Do you run your life? How much of it, and for how long?
P.S. I want better and cheaper models. We’re gonna figure this out like we always do.