Interactive research articles on interpretability, steering, and agent safety. Each one is a paper you can run in the browser.https://seongland.com/en-ushttps://seongland.com/article/asg/https://seongland.com/article/asg/One automaton, built from agent traces, serves as a memory, prediction, and monitoring prior. A compact directly-follows automaton recovered from execution traces unifies workflow memory, next-step prediction, failure prediction, and runtime monitoring across 12 datasets and 8 domains.Mon, 01 Jun 2026 00:00:00 GMTAgentsInterpretabilityAlignmenthttps://seongland.com/article/optimismbench/https://seongland.com/article/optimismbench/Ask a model for the probability something succeeds, then for the probability it fails. The two numbers rarely sum to 100, and the gap leans one way. An interactive look at directional bias in language model probability judgment.Mon, 01 Jun 2026 00:00:00 GMTConfidenceAlignmenthttps://seongland.com/article/paat/https://seongland.com/article/paat/Language agents read every external tool result but are blind to their own internal state. PaaT exposes activation-derived safety probes as an agent-callable read_probe tool — and the same probe flips sign depending on whether the agent chooses to look or the system injects it.Mon, 01 Jun 2026 00:00:00 GMTAgentsInterpretabilityAlignmenthttps://seongland.com/article/agentroom/https://seongland.com/article/agentroom/When agents work the same task, they can't communicate in real time. An interactive look at coordination, not concurrency, as the active ingredient: a real-time coordination room over a CRDT-merged shared workspace.Mon, 01 Jun 2026 00:00:00 GMTAgentshttps://seongland.com/article/confidence-manifold/https://seongland.com/article/confidence-manifold/Interactive exploration of how language models encode confidence — intrinsic dimension analysis, layer evolution, activation steering, and cross-architecture transfer.Sun, 01 Feb 2026 00:00:00 GMTInterpretabilityConfidenceAlignmenthttps://seongland.com/article/crl/https://seongland.com/article/crl/Interactive exploration of Control Reinforcement Learning — learned per-token feature steering with branch analysis, critic trajectories, and layer-wise interpretability.Sun, 01 Feb 2026 00:00:00 GMTSteeringInterpretabilitySAEAlignmenthttps://seongland.com/article/corrsteer/https://seongland.com/article/corrsteer/An interactive exploration of correlation-guided feature selection for controllable language model behavior using Sparse Autoencoders.Fri, 01 Aug 2025 00:00:00 GMTSteeringInterpretabilitySAEAlignmenthttps://www.lesswrong.com/posts/ATsvzF77ZsfWzyTak/dataset-sensitivity-in-feature-matching-and-a-hypothesis-on-1https://www.lesswrong.com/posts/ATsvzF77ZsfWzyTak/dataset-sensitivity-in-feature-matching-and-a-hypothesis-on-1Investigating how dataset composition affects sparse autoencoder feature matching and density patterns.Wed, 01 Jan 2025 00:00:00 GMTSAEDatasetInterpretabilityhttps://seongland.medium.com/reversing-transformer-to-understand-in-context-learning-with-phase-change-feature-dimensionality-13cbf8a2f984https://seongland.medium.com/reversing-transformer-to-understand-in-context-learning-with-phase-change-feature-dimensionality-13cbf8a2f984Understanding in-context learning by reversing transformer representations, exploring phase changes and feature dimensionality.Thu, 01 Aug 2024 00:00:00 GMTInterpretabilityIn-Context LearningTransformerhttps://seongland.medium.com/superposition-hypothesis-for-steering-llm-with-sparse-autoencoder-c07b74d23e96https://seongland.medium.com/superposition-hypothesis-for-steering-llm-with-sparse-autoencoder-c07b74d23e96Exploring how the superposition hypothesis in neural networks relates to steering language models using sparse autoencoders.Mon, 01 Jul 2024 00:00:00 GMTSAESteeringSuperposition