# How I Feel About AI (Zwinkau)

A short, undated post on Andreas Zwinkau's site beza1e1.tuxen.de, one paragraph per feeling and a one-line verdict. It asserts positions rather than arguing them, and its value is as a compact record of how one software developer holds several of them at once.

Surprise comes first, at how well neural networks work. An LLM generates one token after another, like gut feeling, with no planning or reasoning algorithm inside, "yet, some planning and reasoning emerges." This is the same mechanical picture Ted Chiang uses in [[ai-is-not-conscious-chiang]] to argue there is no mind in the machine; Zwinkau starts from the same description and is impressed instead.

Fear: Yudkowsky's argument that a superintelligent AI will inevitably destroy humanity "seems to have no flaw", and nobody seriously sandboxes AI because it is too useful with access.

Disgust at AI companies wrecking the open web. Crawlers and agents descend on open wikis and forums "like a locust plague" and make open communities hard to sustain, the problem [[llm-enshittification]] describes from Gentoo's Bugzilla.

Sadness for artists competing with slop that is cheaper and faster, in a calling that already required sacrifice.

Anger at political systems unable to rein in capital. He would rather a democratic government hold power than billionaires, thinks the problem comes down to making some decisions in a coordinated way, and says most people are not even aware of it. The environmental cost of AI companies consuming resources belongs here too.

Happiness at living in an age of discovery. Software development was "radically transformed forever in 2026", he finds it fascinating to watch, and the tools let him make code, images and music that would not otherwise exist.

His conclusion is that AI seems positive on a technological level and bleak on a societal one. A footer adds that he used Mistral as a reviewer and typed every word himself.

## Where it fits

The piece does not sit in one slot of [[anti-llm-discourse]]. It concedes the harms the sociological writers name and still enjoys the tools, which puts it near the reconciliation writers, [[llm-critics-are-right-use-anyway]] and [[know-thine-enemy]], without their workflow detail or their attempt to justify the position. The footer is a finer-grained [[human-made-disclosure]] than a plain "no AI" line: it separates review from authorship, which is the distinction [[checking-for-human-writing]] argues matters less than who made the judgements.