So what exactly is AI model distillation?
An explainer on distillation, the technique that transfers the abilities of a large 'teacher' model to a smaller 'student' model.
Related stories
NIST curves: P-256 seed was drawn in 1997, Filippo Valsorda checks Anthropic's result
Anthropic found that the unexplained seeds of NIST's P-192 to P-521 curves, supplied by the NSA, came from the lrand48 generator seeded with 1997 and 1998 timestamps. Cryptographer Filippo Valsorda confirmed the finding and puts the remaining freedom at about 30 bits, too little for a backdoor, so distrust of P-256 no longer justifies migrating away from it.
Firms most committed to AI hire up to 25% fewer young people on permanent contracts
A study by economists from HEC Paris, the Banque de France and INSEAD found that about 600 French firms reporting heavy AI use hired 17% to 25% fewer 21-to-29-year-olds on permanent contracts in 2025 than peers in their sector. The researchers saw no clear difference in total headcount.
Inception releases Mercury 3 diffusion LLM at 1,068 tokens per second
Inception has released Mercury 3, a diffusion-based language model it says generates 1,068 tokens per second with a 1-million-token context window. In Inception's own comparison, it beats GPT-6 Luna and Claude Haiku 5.5 on 8 of 12 benchmarks.
Eka is worth $1.45B thanks to a physical AI model that has experts raving
Eka Robotics is now valued at $1.45 billion on the strength of a physical AI model for robotic arms. The company's arm is described as unusually good at picking up a wide range of objects with smooth, human-like movements.
MIT's AI agent society result holds up on consumer hardware
In August, MIT researchers showed that a society of AI agents placed in a shared world without instructions could invent, build and maintain technology better than the same agents working alone. A team working with Apex Fusion has now reproduced the result on consumer hardware rather than a hosted frontier model.