MTR-DuplexBench: Benchmark for Multi-Round Full-Duplex Speech AI
Discover MTR-DuplexBench, a new benchmark for evaluating multi-round conversations in full-duplex speech language models for advanced AI interaction.
Solving the One-Sided Conversation Problem in AI
Explore solutions to the one-sided conversation problem in AI, improving dialogue reconstruction and summary generation from partial transcripts.
Open-Ended Arabic QA Benchmark with Dialect Variants
Explore a new Arabic QA benchmark testing dialects and open-ended questions to improve LLMs' cultural and linguistic understanding.
How Enhancing LLM Reasoning Increases Tool Hallucination
Explore how improving reasoning in large language models can amplify tool hallucination and the trade-offs involved in balancing capability and reliability...
Federated Learning Risks: Dictator Clients Explained
Explore how dictator clients threaten federated learning by disrupting model training and learn strategies to protect decentralized AI systems.