The article details the development of agent-generated CUDA kernels for integrating with H100-based diffusers and transformers models, showcasing benchmarking results and a process for publishing these kernels.

GPT-5.3-Codex-Spark is a new real-time coding model with 15x faster generation and 128k context, currently in research preview for ChatGPT Pro users.

Gemini 3 Deep Think has been upgraded to tackle modern scientific, research, and engineering problems.

The article introduces Expert Pyramid Tuning (EPT), a novel architecture for parameter-efficient fine-tuning in language models that improves task allocation by incorporating multi-scale feature pyramids. It outperforms existing methods on multiple benchmarks while reducing the number of training parameters.

The article tracks the latest AI model updates, including version launches and API changes across major language models like GPT, Claude, Gemini, Llama, with details on new releases such as Nemotron 3 Super and Grok-4.20.

SyGra Studio introduces an interactive environment for generating synthetic data, simplifying the process through a visual interface and direct manipulation of models and prompts.

Allison Murphy describes Axios’ use of AI to assist local reporters, optimize workflows, and produce impactful local news.

The article discusses a metamorphic testing framework for assessing the semantic invariance of Large Language Model (LLM) agents, showing that model size does not correlate with robustness.

The paper presents a reinforcement learning approach for diffusion language models using entropy-guided step selection and intermediate advantages, achieving state-of-the-art results in coding and logical reasoning benchmarks.

The article describes a GitHub repository that tracks knowledge cut-off dates for various large language models, aiding developers in understanding the limitations of these models.