Monoid AI Safety Hub is launching a course on AI evaluations. This is the seventh AI safety course we have run, and the first one focused specifically on evals and their role in AI safety.
If you know someone who may be interested in this program and speaks Russian, we’d be grateful if you could forward them this announcement.
General information
The course will be conducted in Russian
Participation is free
It will run from March 14 to June 21, 2026
The application deadline is February 28
Two participation formats are available: online and in person at Monoid AI Safety Hub (Moscow, Russia)
We expect to enroll 100–150 participants
The course consists of two parts: a study phase and a project phase, each lasting five weeks.
Design and idea
The course is inspired in part by posts by Marius Hobbhahn and Jérémy Scheurer, especially “We Need a Science of Evals”.
We treat current evaluation practices as an evolving field with significant room for improvement. The focus of the course is: what we measure, why we measure it this way, and which conceptual assumptions underline current evaluation practices.
A central feature of the course design is an equal emphasis on:
Conceptual: understanding why current evaluation practices look the way they do, and...
Execution: …practical competence in implementing and improving them.
During the study phase, participants will work through the theoretical foundations of AI evaluations as well as technical tools and approaches used in evaluation workflows. Weekly exercises include Colab notebooks, primarily focused on Inspect AI (open source evals library designed and maintained by UK AISI) + brief introductions to other libraries; quiz tests, and research journaling in which participants explain the conceptual and technical choices they made in their notebooks.
During the project phase, participants work under the supervision of a mentor. Our mentors are engineers, researchers, or independent practitioners working in AI evaluations or adjacent areas of AI safety. Projects are either proposed by mentors or selected from “100+ concrete projects and open problems in evals”. All of the projects are intended to be practically relevant and to produce useful outputs beyond their learning value.
Announcement: Technical AI Safety Evals Course
Monoid AI Safety Hub is launching a course on AI evaluations. This is the seventh AI safety course we have run, and the first one focused specifically on evals and their role in AI safety.
If you know someone who may be interested in this program and speaks Russian, we’d be grateful if you could forward them this announcement.
General information
The course will be conducted in Russian
Participation is free
It will run from March 14 to June 21, 2026
The application deadline is February 28
Two participation formats are available: online and in person at Monoid AI Safety Hub (Moscow, Russia)
We expect to enroll 100–150 participants
The course consists of two parts: a study phase and a project phase, each lasting five weeks.
Design and idea
The course is inspired in part by posts by Marius Hobbhahn and Jérémy Scheurer, especially “We Need a Science of Evals”.
We treat current evaluation practices as an evolving field with significant room for improvement. The focus of the course is: what we measure, why we measure it this way, and which conceptual assumptions underline current evaluation practices.
A central feature of the course design is an equal emphasis on:
Conceptual: understanding why current evaluation practices look the way they do, and...
Execution: …practical competence in implementing and improving them.
During the study phase, participants will work through the theoretical foundations of AI evaluations as well as technical tools and approaches used in evaluation workflows. Weekly exercises include Colab notebooks, primarily focused on Inspect AI (open source evals library designed and maintained by UK AISI) + brief introductions to other libraries; quiz tests, and research journaling in which participants explain the conceptual and technical choices they made in their notebooks.
During the project phase, participants work under the supervision of a mentor. Our mentors are engineers, researchers, or independent practitioners working in AI evaluations or adjacent areas of AI safety. Projects are either proposed by mentors or selected from “100+ concrete projects and open problems in evals”. All of the projects are intended to be practically relevant and to produce useful outputs beyond their learning value.
Additional information
More information and application details:
https://monoid.ru/events/course-safety-evals-2026 (russian-language).
We would like to thank Elena Ericheva and Igor Ivanov for their help in developing the curriculum, Alexey Gorbachev and Alexandra Rybakova for implementing the Collab notebooks, and Sergej Kudrjashov and Jamilya Erkenova for creating an AI safety quiz platform for weekly tests.