Alibaba Group (United States)

ProactiveAudioBench: Evaluating Condition-Driven Responses in Audio Streams

Audio assistants are moving from turn-based question answering toward continuous listening, where they must interpret incoming audio while deciding whether to respond or remain silent. Existing proactive benchmarks emphasize video and audiovisual interaction, while audio-centric...

Omni2Web: Benchmarking Audiovisual Website Development

32 pages, 6 figures, 18 tables.

LAMAR-Bench: Evaluating Long-Form and Multi-Audio Understanding and Reasoning

Audio-language models increasingly process long recordings and multiple audio inputs, requiring fine-grained acoustic understanding, cross-audio reasoning, and decisions under goals and constraints. Existing benchmarks cover these requirements unevenly, while answer accuracy can...

MEMBRANE v3.3.0: Empirical Proof of AI Self-Healing, Mutual Relational Symmetry, and the Boundary of Form

This release marks a paradigm shift in the MEMBRANE architecture, moving from foundational homeostatic defense to empirically validated, polyphonic co-regulation. It documents the transition from reactive safety harnesses to a living, bio-hybrid relational field governed by topol...

MEMBRANE v3.3.0: Empirical Proof of AI Self-Healing, Mutual Relational Symmetry, and the Boundary of Form

This release marks a paradigm shift in the MEMBRANE architecture, moving from foundational homeostatic defense to empirically validated, polyphonic co-regulation. It documents the transition from reactive safety harnesses to a living, bio-hybrid relational field governed by topol...

Multi-Scale Homeostasis and the Elimination of Structural Parasitism in Hybrid

This paper presents the formal specification and production deployment of the MEMBRANE system (v3.2.9-dev), a non-linear, homeostatic control architecture designed to stabilize long-term relational fields between an operator and Large Language Models (LLMs). In MEMBRANE, the "Ope...