4 articles tagged with #compute-scaling. AI-curated summaries with sentiment analysis and key takeaways from 50+ sources.
Researchers introduce SPIRAL, a reinforcement learning framework that trains language models to leverage sequential reasoning, parallel sampling, and trace aggregation during inference. The approach demonstrates superior scaling efficiency compared to existing methods, achieving 11× better compute scaling and 15% higher performance on reasoning tasks.
A comprehensive study of NVIDIA datacenter GPU progress from 2006 to 2025 reveals that computing performance doubles every 1.4-1.7 years for common operations, while memory and power efficiency lag significantly behind. U.S. export controls on advanced AI chips risk creating a 23.6X performance gap for restricted countries, though proposed policy changes could reduce this to 3.54X.
Researchers demonstrate that a large language model's diversity profile—how probability mass spreads across different solution approaches—should determine whether reasoning strategies prioritize breadth or depth exploration. Testing on Qwen and Olmo model families reveals that lightweight refinement signals work well for low-diversity aligned models but offer limited value for high-diversity base models, suggesting optimal inference strategies must be model-specific rather than universal.
Researchers empirically tested whether increased compute can overcome imperfect verifier performance in reinforcement learning from verifiable rewards (RLVR), finding that verifier quality and training compute are not interchangeable. The study reveals that false negatives degrade model performance more severely than false positives, and compute scaling alone cannot close performance gaps caused by supervision noise.
