Better_AI_Meetup on Deep Thinking: Reinforcement Learning and Benchmarking for Better LLMs

Join Better_AI_Meetup, a space for AI designers, builders, and everyone who believes AI can make a positive difference.

What actually makes a language model better? Is it more data, a bigger model, or something else entirely? At the 20th Better AI Meetup, we’ll look at what happens after a model is trained, and why careful evaluation matters just as much as the model itself.

Jakub Mačina, a researcher at ETH Zurich, will share how reinforcement learning can help language models reason and collaborate more effectively with people. He’ll also show how a thoughtfully trained 7B open model can keep up with much larger proprietary models.

In the second talk, Marek Šuppa from Slido and Andrej Ridzik from KInIT will explore how we measure the real capabilities of language models. They’ll talk about why no single benchmark can tell the whole story, what makes evaluating Slovak LLMs particularly challenging, and how benchmarking helps separate genuine progress from marketing claims.

5:00 pm_Introduction

5:05 pm_Jakub Mačina, Researcher at the ETH Zurich

5:30 pm_Marek Šuppa, Principal Machine Learning / AI Engineer at Slido, _Andrej Ridzik, Senior Research Engineer at KInIT

5:55 pm_Q&A

6:30 pm_Networking

As usual, all talks and the discussion will be in English. Find out more about the Better_AI_Meetup here.