EPFL

Apertus ships in 8B and 70B sizes, claims more than 1,000 languages, and bakes in EU-AI-Act compliance: PII removal, opt-out respect, memorization prevention. The pitch is total reproducibility — "to AI as Open is to Source" — not benchmarks. Cutoff: March 2024.

One founder reports building real RAG software on it today; it is not ready for agentic work, but it drives. Skeptics aimed at the institution: a model moving at committee speed, a country with no domestic GPUs, and a project whose best output may be the people it trained.

Neural cellular automata are bio-inspired: identical cells run a learned local rule and self-organize into an image, with the regeneration tricks of living tissue. The catch was always resolution — compute scaled quadratically with the grid.

The EPFL/Google fix pairs a coarse NCA grid with a tiny implicit decoder that fills in arbitrary-resolution pixels. The demos are live and abusable: a few brush swipes, one commenter found, and the stabilized kitten dissolves completely.