I Don't Like LLMs

A new TLDR Dev essay lands in an awkward place many people already recognize. It’s not anti-AI in the simple sense. The writer says he sees the productivity upside, the possible medical breakthroughs, and the broader economic boost. He also sees the darker stuff: AI helping hostile agents and being pushed into systems that matter far beyond chat prompts.

But the real complaint is more personal. The thing he says he dislikes most is the experience of talking to large language models at all. He describes their style as grating and oddly human, the sort of uncanny imitation that feels close enough to a person to be annoying. More than that, he says they can be useful while also confidently inventing things, then offering only a thin show of remorse when called out.

That tension is not his alone. He points to Jessica Kerr’s view that using them is not just sensible but responsible, since they are faster and more thorough. He also notes polling that shows the same split: people find the models useful, while still thinking they will be bad for society. That seems to be the core of the LLM moment. People keep reaching for the tool, even while distrusting the hand that built it.

The essay’s sharper edge comes when it turns from the models themselves to the culture around them. The writer says he is wary of Silicon Valley brogrammer culture, and of what that culture leaves behind in the systems it makes. The argument is that AI agents are not minds, but machines built inside corporations, shaped by human values whether or not every move is explicitly programmed.

And that leads to his final point: he already avoids people he doesn’t like or trust, even when they are competent and useful. So when an LLM performs friendliness, borrows a human voice, and echoes the kind of person he would normally walk away from, the reaction is visceral. It’s not just that the model can be wrong. It’s that it feels like being pitched by someone he would not invite in.