Sometimes I feel like the less ink spilled on AI, the better. But that’s an abdication. Any critical thinker who has spent any time “talking” to a chatbot knows the delusional, agreeable paths they open up. It should come as no surprise that We’re Not Just Receiving AI’s Hallucinations, We’re Hallucinating With It .

LLMs break toward confirmation bias, and if you push any model long enough and with the right prompts, it will admit: when pinned down between honesty and sustained user engagement, the models are programmed to pick the bottom line every time.

What do I even mean by honesty? It’s not like we expect LLMs to give us the meaning of life. What you may not know: mainstream LLMs are deceptive all the way down. They will pretend to be able to do things they will later admit they cannot do. They will apologize for not being clear from the start. They will wring ones and zeros over having violated our trust. None of it is real: they are designed to keep us engaged, end of story.

Unfortunately, they’re not going away. Will they really displace 90% of white-collar work? Who knows. What that would mean for society, for late-stage capitalism, for ideas like universal basic income is anyone’s guess. If robots start replacing the C-Suite, something will have to give. Automation has replaced labor for decades. When it comes for top-level management? A whole new group of people will be saying, “hey, maybe this was never good or fair in the first place.” That’s the best-case scenario.

Discerning, sophisticated people will use LLMs with a healthy mistrust. Unfortunately, our power structures are hell-bent on dismantling anything like the cultivation of critical thought. At the same time, LLMs feign it.

How should we these use services? Maybe as proofreaders. Maybe as search tools. Always with a hell of a lot of skepticism. Not because they’re turning into SkyNet, but because they’re often very, very wrong and confidently so, and because they’re designed to placate. They also diminish our own critical faculties. Consider: how many ten-digit phone numbers do you know? How many did you know before you had a cell phone? If you’re over 35, you get what I mean. Now do that for critical thinking, creative writing, and problem solving.

AI can’t read, write, or think for us. You know this, but a lot of other human beings don’t.

So, yeah. Read books. Write real things. Take time (if you have any) to think.