LLMs won, and I'm not happy about it

2026-07-28 00:00:00 -0500

LLM: a type of artificial intelligence that can understand and generate human language, e.g. ChatGPT

Sam Altman was right. The best interface for computers is natural language. They are a big deal. LLMs won, and they were always going to. I’m not happy about it.

The public has received LLM chatbots and they love it. It’s the chat element, not the training, trillions of parameters, or reasoning capabilities, that makes the difference for adoption. People don’t like one-way communication. I’ve seen it first-hand when a non-technical person goes through any amount of difficulty, even when met with plain-language error messages. They freeze up.

IT people know this. Even if the message is plain-as-day, they will get confused, like it’s in another language. And anyone who’s helped someone with a problem—or probably been the one asking—knows that they’ll explain the whole issue, start-to-finish. The technician has to pick apart what’s actually important, what needs to be done, etc.

Web forms can’t do this. Search engines can’t do this. LLMs can do this right now. You can write an initial prompt going over every single detail of a problem and it will identify action items. In fact, it’s the best way to use an LLM chatbot rather than sprinkling in new information as-you-go. The best part? If someone doesn’t give enough information, the chatbot will almost always ask for it. This is what non-technical people NEED. They don’t know what they don’t know. I am not going to pretend like I’ve never gone to an LLM for technical assistance—I do new things all the time, and I’ve only got so many hours in the day. Thing is, these days they’re actually helpful. And if it’s a topic I’m unfamiliar with, I can usually get acquainted with a cursory web search first.

Okay, you can stop booing me. Here’s the hangup. I don’t want to talk with a computer most of the time. There are times when I might. But that has to be on MY terms. If I ever get redirected to, say, an LLM support bot or call, I’m going to be upset. If the kiosk starts talking to me, I’m going to be upset. And my god, the machine better not start lecturing me, as if it has opinions and agency. It’s like this meme, except actually real:

Sam Altman and the other AI giants were wrong about one thing, at least. People do not want to exclusively use chat interfaces. If they did, then smartphone assistants would’ve won, too. They’re still mostly relegated to sending text messages in the car. This is why Copilot in Windows was doomed to fail. Microsoft wanted to replace people’s bread-and-butter workflows with chat interfaces, to make an “agentic OS”. Now the second search auto-complete for “windows copilot” is “windows copilot disable”. This is the kiosk failure. For the simple and routine, making a few clicks or taps will beat the conversation every single time. Forever. No thanks machine, I’ll just do it myself.

(Obvious disclaimer: nothing on this site is AI-generated. If it was, it’d probably look better.)