Much of the evaluation and tuning of a search system relies on relevance labels—annotations that say whether a document is useful for a given search and searcher. Ideally these come from real searchers, but it is hard to collect this data at scale, so typical…
The rapid progress in Large Language Models (LLMs) poses potential risks such as generating unethical content. Assessing LLMs' values can help expose their misalignment, but relies on reference-free evaluators, e.g., fine-tuned LLMs or close-source ones like GPT-4, to identify values reflected in generated responses. Nevertheless,…
AI for Domains (AID) Our mission The M365 Research organization's AI for Domains (AID) team is an applied research group focused on leveraging cutting-edge research to extend the capabilities, efficiency, reliability, and reduce risks associated with Copilot. We prioritize user privacy and confidentiality while unlocking…
Much of recent progress for natural language generation (NLG) has been in the context of English and, in general, high resource languages, however, Indian languages have yet to see similar paradigm shifts despite their speaking population comprising about a fifth of the world's population. Two…
A public framework for time-series forecasting with spiking neural networks (SNNs).
Software engineering activities such as package migration, fixing error reports from static analysis or testing, and adding type annotations or other specifications to a codebase, involve pervasively editing the entire repository of code. We formulate these activities as repository-level coding tasks. Recent tools like GitHub…
The Intelligence Toolkit is a suite of interactive workflows for creating AI intelligence reports from real-world data sources. The toolkit is designed to help users identify patterns, answers, relationships, and risks within complex datasets, with generative AI (OpenAI GPT models) used to create reports on…
Printed circuit boards are abundant—in the stuff we use and in landfills. Researcher Jake Smith and professor Aniruddh Vashisth discuss the development of vitrimer-based PCBs that perform comparably to traditional PCBs but have less environmental impact.
VentureBeat announced the winners of the sixth annual Women in AI Awards yesterday at VB Transform. The "Rising Star" award honors a woman in the early stage of her AI career who has demonstrated exemplary leadership traits. Amini focuses on engineering new technologies for precision…
We present MELLE, a novel continuous-valued tokens based language modeling approach for text to speech synthesis (TTS). MELLE autoregressively generates continuous mel-spectrogram frames directly from text condition, bypassing the need for vector quantization, which are originally designed for audio compression and sacrifice fidelity compared to…
When Large Language Models (LLMs) are compressed using techniques such as quantization, the predominant way to demonstrate the validity of such techniques is by measuring the model's accuracy on various benchmarks.If the accuracies of the baseline model and the compressed model are close, it is…
