Choose a note to see what it builds on, where it is used, and which notes it references.
235 notes
Understand coding-agent execution, recovery, and acceptance through harness responsibilities, Pi's minimal core, and its safety boundaries.
How step limits, tool validation, verification, cancellation, and durable state keep agent loops useful and recoverable.
How multi-tiered memory and AST-level retrieval align agent intent, compress token overhead, and prevent duplicate generation.
Let AI find gaps, explain one step, and provide practice while keeping recall and problem solving with the learner.
Understand production agent harness responsibilities through Codex runtime state, context compaction, integration surfaces, and two-layer security controls.
How to assemble a small, current working set while preserving provenance and durable state outside the model window.
A practical way to label specifications, vendor claims, benchmark results, observations, and recommendations without pretending they prove the same thing.
A practical map of the control, context, tool, evidence, and evaluation layers around AI agents.
A protocol-level view of MCP architecture, lifecycle, primitives, transports, trust boundaries, and alternatives.
How to package a repeatable agent procedure, test its trigger and outcome, and manage portability and code-execution risk.
How classical software engineering principles—design concepts, ubiquitous language, TDD, deep modules, and gray-box delegation—resolve failure modes in AI-assisted coding.
Design tools whose inputs, authority, side effects, failures, and evidence remain legible to models and operators.
Block target, group, temporal, and preprocessing leakage by defining prediction time, entities, and train-only pipelines.
Start from deployment distribution, splits, metrics, thresholds, and uncertainty instead of treating one test score as universal ability.
A map from data-generating processes and provenance to leakage-resistant splits, distribution-shift evaluation, and authoritative public sources.
A path from questions and data-generating processes to reproducible transformations, leakage-resistant splits, distribution-shift evaluation, and bounded conclusions.
A compact vocabulary for discussing datasets, modeling choices, evaluation, and reproducibility without product-specific jargon.
A curated entry to legacy machine-learning, deep-learning, and data notes, with explicit coverage gaps and promotion rules.
Distinguish ReLU, GELU, SiLU/Swish, and GLU variants, then select nonlinearities with controlled modern-network experiments.
How MHA, GQA, MLA, cache quantization, and token-selection methods trade memory for implementation and quality risk.
Understand content addressing, Q/K/V, masks, multi-head variants, efficient implementations, and interpretation limits through a numerical example.
Convolutional inductive bias, channels, padding, stride, pooling, and the distinction between translation equivariance and invariance.
The linear prediction model, squared-error objective, solution methods, and the assumptions that determine what its coefficients mean.
Understand MLP capacity, backpropagation, optimization failures, and inductive bias through tensor shapes and a worked XOR construction.