Efficient Training of Diffusion Mixture-of-Experts Models: A Practical Recipe

Mistral AI has launched the Mistral 3 family, a suite of 10 open-source models designed for diverse applications, including smartphones, drones, and enterprise systems. This release represents a significant advancement in Mistral's efforts to compete with major tech players like OpenAI and Google, as well as emerging competitors from China.

Researchers have found that large language models (LLMs) make decisions about networking and friendship in ways that closely resemble human behavior, both in synthetic simulations and real-world contexts. This suggests that LLMs can replicate social decision-making processes similar to those of people.

DeepSeek is set to launch its new AI model, R2, after delays attributed to limited access to computing resources. This development is expected to heighten the ongoing competition between the U.S. and China in the artificial intelligence sector.

DeepSeek has made headlines again with its recent advancements in artificial intelligence, particularly with the introduction of its new model, DeepSeekMath-V2, which has achieved gold medal status at both the International Mathematical Olympiad 2025 and the Chinese Mathematical Olympiad 2024. This achievement highlights the company's growing influence in the AI sector.

A recent study introduced Semantically Equivalent and Coherent Attacks (SECA) aimed at eliciting hallucinations in Large Language Models (LLMs) through realistic prompt modifications that maintain semantic coherence. This approach addresses the limitations of previous adversarial attacks that often resulted in unrealistic prompts, thereby enhancing the understanding of how LLMs can produce hallucinations in practical applications.

A recent study has demonstrated a method for predicting the performance of black-box language models (LLMs) by utilizing follow-up queries to assess their outputs. This approach allows researchers to train reliable predictors based on the probabilities of responses, achieving accuracy that can surpass traditional white-box models that analyze internal mechanisms.

A new Conformer-based decoder has been developed for the LibriBrain 2025 PNPL competition, focusing on Speech Detection and Phoneme Classification using 306-channel MEG signals. The approach includes a lightweight convolutional projection layer and task-specific heads, achieving notable performance with 88.9% accuracy in Speech Detection and 65.8% in Phoneme Classification, ranking in the top-10 for both tasks.

A new quantization method called Four Over Six (4/6) has been introduced to enhance the NVFP4 quantization algorithm, which is crucial for large language models (LLMs). This method evaluates two potential scale factors for each block of values, addressing issues of performance degradation during inference and divergence during training that arise from quantization errors in floating-point formats.