Research
158d agoSpecialized Language Models from Mixed Pretraining
Researchers at Apple Machine Learning Research propose a new method for dividing large language models into domain‑specific specialists, reducing cross‑domain interference while preserving overall knowledge.
The Signal
The technique, showcased at ICLR 2026, enables faster, more accurate deployment in fields ranging from legal to medical translation, promising global improvements in AI accessibility and performance.