Insights from Big Techday 26: Two talks by Prime Intellect

At our recent Big Techday 26, we welcomed two speakers from Prime Intellect: Konstantin Dunas and Florian Brand. Their talks are now available on our YouTube channel:

🔹In the first talk “Scaling LLM-RL for the age of agents”, Konstantin discussed how LLM agents changed software engineering workflows, and why reinforcement learning for language models required a different perspective than classical RL. He also shared insights into the infrastructure challenges behind agentic training, from coordinating inference and training nodes to designing robust environments for long-horizon tasks.
Watch the full talk here.

🔹In the second talk “LLM benchmarks in the time of agents”, Florian examined how LLM evaluation has evolved from knowledge-based benchmarks to reasoning and agentic tasks. He showed why benchmark results should be interpreted with care, as prompts, harnesses, sampling parameters, and infrastructure choices could significantly influence the reported scores.
Find the recording here.

Many thanks to both speakers for sharing their insights and perspectives about where LLM agents, reinforcement learning, and the evaluation of AI models are heading.