Alex Volkov hosts and curates ThursdAI, a weekly show about AI research, open models and developer tools. He is an AI evangelist at Weights & Biases and publishes practical evaluation materials.[1][2][3]
Start with LLM Evaluations Workshop →
X Career
Volkov has run ThursdAI since 2023, beginning with a live X Space. The show now combines live discussions, an edited podcast and a newsletter with source links. Researchers, engineers and founders discuss model releases and their implications for people building applications. His host biography also identifies him as CEO of Targum.video, an AI video-translation company.[1]
Contributions
Volkov’s LLM Evaluations Workshop uses Bluesky posts as classification data. Participants trace model calls, create and annotate datasets, and compare outputs through W&B Weave. The example connects an interactive interface to evaluation records, so a team can inspect the actual cases behind a score.[3]
Ideas
The workshop distinguishes programmatic scoring for structured outputs, human annotation and model-based judging for more open-ended responses. Each method serves a different role: exact checks enforce known formats, people establish reference labels, and another model can assist with judgments that are harder to encode. The materials include limitations rather than treating an automated judge as a complete replacement for evaluation design.[3]
Tools, writing & talks
Tools
Start here
LLM Evaluations Workshop
github.comA hands-on classification example connecting traces, datasets, human labels and model-based evaluation.
Start with the workshop’s evaluation approaches, then its notebook. Running the exercises requires Python and the documented provider credentials; reading the design requires only basic familiarity with model outputs.
Author, created for Weights & Biases[3]
Writing, research & talks
A group discussion about which AI products and workflows remain useful in daily work.
Participant[4]Original source