ReflectionAI's Beam: 500B open-weight model challenges closed AI future

ReflectionAI co-founder and CEO Misha Laskin joined the No Priors podcast to discuss Beam, a 500 billion parameter open-weight reasoning model the company recently launched. He explained the pre-training and reinforcement learning approaches that enable Beam's reasoning efficiency, and argued that open models will capture the majority of global token demand as enterprises shift from renting AI capabilities to owning their own intelligence.

Laskin also addressed competition with Chinese open-source models, emphasizing that global competition drives innovation rather than threatening US interests. He discussed safety considerations around releasing frontier-level open models and their potential to accelerate scientific discovery by giving researchers direct access to powerful reasoning tools.

The conversation covered business models for open-weight AI, data center economics, and the ratio of research headcount to compute investment as the field scales.