Import AI 476: Creativity and RSI; worse-than-China open weight models; AI goes to space

The title lists three subjects alongside the model coverage: creativity and what is abbreviated as RSI, open weight models, and AI in space. Two further items named in the excerpt are an AI startup called Reflection and a benchmark called TasteVal. No other organisations, individuals or figures appear in the excerpt.

Reflection has revealed a model named Beam. Beam is described as a 501 billion parameter mixture-of-experts model with 23 billion active parameters, and it is being promoted on the basis of token efficiency. The excerpt does not state what the remaining parameters do, nor when Beam was released. The issue also reports on TasteVal, described as a benchmark, which found that the experimental research taste of frontier models has doubled every 3.0 months since December 2025. The excerpt does not define how 'research taste' is measured.

The newsletter's title also references 'creativity and RSI', open weight models characterised as 'worse-than-China', and an item on AI in space. The excerpt does not expand the abbreviation RSI, nor does it state what the space-related item concerns. It also provides no further detail on the open weight models described as worse than China, or on any other coverage contained in issue 476.