Firehose

This paper investigates whether large language models, like Google's Gemma-4-E4B-it, represent scientific concepts and governing physics, and whether this representation affects their answers. Practitioners caring about the accuracy and reliability of language models in scientific domains might find this research valuable.

The Gemini 3.6 Flash model is now available, offering stronger performance on complex agentic and multimodal tasks, reduced token usage, and lower pricing compared to the 3.5 Flash model. Additionally, the 3.5 Flash-Lite model, the fastest and lowest-cost model in the 3.5 family, has been updated with improved performance for high-throughput execution. These models are now the recommended choice for access to the latest features and models, replacing the deprecated temperature, top_p, and top_k parameters. AI summary

This paper investigates how Diffusion Language Models (DLMs) internally represent time and how this representation can be used to modulate the model's behavior. Practitioners might care because understanding how DLMs process time could lead to more controllable and interpretable models.

Tim Scarfe interviews the Tufa Labs ARC-AGI-3 team to dissect their winning approach on the ARC-AGI-3 benchmark, focusing on how their system discovers goals and balances exploration with action efficiency. The episode explores the challeng…