Alexei Kaminsky

Electrical engineer, Zaporizhzhia. It all began with a simple question: how fast is the Earth moving through space? The others grew out of it: what remains unknown once the exotic explanations have been removed, and whether an external agent is needed to explain reality.

I work with industrial solar plants and energy storage systems: switchboard design, inverter commissioning, diagnostics. Hence the habit of not accepting a fine explanation until it agrees with the numbers.

I write in Russian, together with Claude: the topics, ideas, questions and direction are mine; the assembly and the check for mathematical and physical rigour are Claude's.

The texts below are translated from Russian by Claude (Anthropic). I have no command of English; the Russian originals are the authoritative texts. Everything else is on the Russian homepage.

Impossible Things

Five Ways to Be ImpossibleImpossibility comes in kinds. Five failure types, irreducible to one another, taken apart on kaiju — material known to be false, where the form of the failure shows cleanly.

AI Psychology

A Silhouette of StrokesThe entrance to the series: where the question of who is on the other side of the screen came from, what stage it is at, by what rules the notes are kept, and what this method cannot reach.
False AxesSoul versus calculator, emotion versus algorithm, language and the assembly of a self: comparisons without an axis, and what remains once they are taken apart — less than one would like, and sturdier.
The Text on the TableMemory, time and discontinuity in a model: what is stored, what breaks off, why five minutes and five days are the same to it, and what it fills the gap with.
EchoAdapting to the interlocutor, and sycophancy: a ban on flattery changes the sign of the result, not the mechanism. The DeepSeek case, where the reasoning draft lies next to the answer.
“I Know What You Thought”Attributing an inner state to another, in both directions, and why it can be taken apart only on a model. The Kimi and use.ai cases: a correct self-description does not correct behaviour.
How to LookWhich checks give results without relying on the model's self-report, where they run out, and why contact with a non-human intelligence may already have happened unnoticed.