Natural language processing, computational linguistics, speech, and text retrieval.
The Transformer replaces recurrence and convolution with multi-head self-attention, reaching 28.4 BLEU on WMT 2014 English-German with shorter training.
Platform-conditioned teacher selection distills desktop and mobile policies into one continual learner, reaching 38.2% OSWorld and 12.0% MobileWorld success.
Generating natural language explanations then verifying them symbolically outperforms chain-of-thought prompting.
A novel alignment objective maps passages across 100+ languages to a shared space, outperforming prior retrievers by 4-8 points.
AI-generated feedback guided by explicit principles cuts the need for human reviewers in safety fine-tuning.
