AI Weekly Malaysia

Latent Space interviews Fabian Theis of Helmholtz Munich about pushing biology from mapping and predicting toward controlling it, using multi-scale, multi-modal, time-varying 'multiomics' data. The episode contrasts heavy attention on protein/molecular design with quieter skepticism about Virtual Cell models, which still face questions about solving basic problems. It also cites a $1.8B Biohub/NIH/DOE/Google/Meta commitment tied to new imaging and measurement capabilities that generate large datasets for AI training.

Why: There is no direct API, tooling, pricing, or Malaysia/SEA policy detail in the excerpt, so most builders should not change course from this alone. The actionable signal is for AI/ML learners tracking AI-bio: watch whether the $1.8B commitments actually release usable imaging/multiomics datasets and benchmarks, since the text says Virtual Cell models still face basic effectiveness questions.