Microsoft's new AI training method eliminates bloated system prompts without sacrificing model performance

Microsoft researchers have developed On-Policy Context Distillation (OPCD), a training method that permanently embeds ...

New computation method for climate extremes: Researchers reveal 10-fold increase in heat over Europe

How much will heat, flooding, drought and storms increase as a result of human-induced climate change? In a groundbreaking ...

Predicting extreme rainfall through novel spatial modeling

Japan is an archipelago with diverse climate zones and complex topography that is prone to heavy rain and flooding. Add the ...
Nature

Computational methodology for drug discovery

Despite significant progress in medicinal chemistry and life sciences research, drug discovery and development remain slow and expensive, taking on average approximately 15 years and US$2 billion to ...

Researchers baked 3x inference speedups directly into LLM weights — without speculative decoding

Researchers from the University of Maryland, Lawrence Livermore, Columbia and TogetherAI have developed a training technique that triples LLM inference speed without auxiliary models or infrastructure ...
Nature

Ultra-sensitive CAR T cells eliminate hard-to-treat tumours in mice

A proof-of-concept study opens up an avenue for treating solid tumours that express low levels of a target antigen.