ArXiv paper providing methodology for auditing the reliability and trustworthiness of concept-based explainability methods in AI models.
Research on sparse attention kernel optimizations enabling efficient training of models on million-token sequences. Addresses computational bottleneck in long-context model training.
PrismML, a Khosla-backed startup, released compressed version of Qwen-3.6-27B capable of running on iPhone devices. Claims breakthrough in largest AI model deployable on mobile.
ArXiv paper on autonomous AI agents for discovering and exploiting IoT vulnerabilities. Addresses critical cybersecurity challenges in connected device ecosystems.
Research paper on scalable visual pretraining methods that enhance language model capabilities. Received 25 HuggingFace upvotes.