True Positive Weekly #140
The most important artificial intelligence and machine learning news and articles
Hey, in this issue: top 5 AI model optimization techniques; accelerating scientific discovery with AI; validating LLM-as-a-Judge systems; measuring thinking efficiency in reasoning models; the rise of subagents; open agentic coding; Google’s efficient embedding model; and more.
[Nvidia] Top 5 AI model optimization techniques for faster, smarter inference
[Google] Accelerating scientific discovery with AI-powered empirical software
[Google] AlphaEvolve on Google Cloud: AI for agentic discovery and optimization
[CMU] Validating LLM-as-a-Judge systems under rating indeterminacy
Measuring thinking efficiency in reasoning models: The missing benchmark
[ArXiv] DeepCode: Open agentic coding
[Model] Welcome EmbeddingGemma, Google’s efficient embedding model
Enjoy the newsletter? Please, don’t forget to like or comment to help the newsletter grow.
Great curation this week. The pieces on measuring thinking efficiency in reasoning models and the rise of subagents are particuarly timely, feels like we're seeing a shift from monolithic models to more modular, task-specific agents. The LLM-as-a-Judge validation work from CMU is important too, lots of people are relying on tht approach without understanding rating indeterminacy.