Zhipu AI has open-sourced GLM-ASR-Nano, a 1.5B parameter speech recognition model designed for robustness in real-world scenarios. It achieves an average error rate of 4.10, significantly outperforming Whisper V3 (6.93) on open-source benchmarks, with exceptional performance in dialects, low-volume speech, and code-switching. The model utilizes an asymmetric 'heavy perception, extreme compression' architecture and is already deployed in Zhipu's production input method.
Robustness
鲁棒性
A model's stability when facing noise, interference, or out-of-distribution data
Curated Intelligence
A Google team published a Nature study showing that reorganizing internal visual representation spaces makes AI perceive the world more like humans. This approach significantly improves multimodal models' robustness, reliability, and cross-concept generalization, addressing the failure of traditional vision models to capture high-level semantic commonalities (e.g., cars and airplanes as large metal vehicles).
OpenAI has open-sourced Whisper, a neural network that achieves human-level robustness and accuracy in English speech recognition. This release marks a significant breakthrough in high-precision, multilingual transcription capabilities, providing developers with a powerful open-source tool.
Researchers introduced the Unforeseen Attack Robustness (UAR) metric to evaluate how reliably neural network classifiers defend against adversarial attacks unseen during training. The method highlights the necessity of measuring performance across a diverse range of unforeseen attacks, addressing limitations in current single-benchmark evaluations.
Researchers created images that reliably fool neural network classifiers across varied scales and perspectives. This finding directly challenges recent claims that multi-view data collection makes autonomous vehicles resistant to malicious adversarial attacks.