1. Research PapersScore77

    Statistical Self-Consistency in Language Models

    A new research paper explores how well Large Language Models (LLMs) adhere to basic probabilistic principles, specifically the law of total probability, when performing in-context learning. The study found widespread violations of these consistency properties across various models and domains.

    Source: Partition, Prompt, Aggregate: Statistical Self-Consistency in Language Models (arxiv.org/abs/2607.15277v1) Full analysis
  2. RoboticsScore72

    RoboTTT: Scaling Robot Policy Context to 8K Timesteps

    Researchers have introduced RoboTTT, a novel robot model and training methodology that significantly expands visuomotor context to 8,000 timesteps. This advancement enables new capabilities such as one-shot imitation learning and improved performance on long-horizon tasks without increasing inference latency.

    Source: arXiv research paper 'RoboTTT: Context Scaling for Robot Policies' by Yunfan Jiang et al. Full analysis