Igor Kapustin

The Lesson Path. How a person's correction becomes a verified lesson for an AI model. Open specification, version 0.1

An open specification (version 0.1, a draft for discussion) describing how a correction a person makes while working with an AI model can become a verified lesson for the next version of the model — and be credited to its author. Seven stages: free thinking (not used for training...

Second-Generation AI: The Hybrid Survivability Criterion. Open specification, version 0.2 / ИИ второго поколения: критерий гибридной выживаемости

An open specification proposing that a generation of AI be defined not by a model's power or speed, but by the hybrid survivability of the human operator and the machine as a single system: its harmlessness in three circles — for the operator, for the people around while it works...

The Blind Kittens and the Trainers of Their New AI Models: Whose AI will be better — one trained by people who use AI every day, or by people forbidden to use it?

An essay on how the AI industry selects the people who train and evaluate its models. Platforms that recruit AI trainers usually forbid them to use AI tools in their tasks, while the labs themselves bring AI assistants into their own evaluation work. The essay argues that such a...

Conditions of Admission and Participation for People Testing and Raising AI Models. Specification v0.1

An open specification of conditions of admission and participation for people who test, evaluate and raise AI models. It covers declaration instead of prohibition of AI tools, a field for observations that cannot yet be named, deferred reports on what came of a decision, payment...