-
Fix VS Code Remote-SSH Repeated Verification Prompts with 2 Factor Authentication
A short fix for repeated 2FA verification prompts when VS Code Remote-SSH installs VS Code Server on a new remote machine.
-
Asymmetric Risks of Unfaithful Reasoning: Omission as the Critical Failure Mode for AI Monitoring
A framework for thinking about faithful reasoning, monitorability, and why omitting causal factors can be more dangerous than adding irrelevant ones.
-
Efficient Streaming Language Models with Attention Sink - Xiao et al.
Notes on attention sinks, streaming language models, and the empirical behavior of window attention.
-
Debugging Numeric Comparisons in LLMs
Layerwise geometry shows the model internally separates Yes/No, but a last-layer readout corrupts the decision—especially for decimals.