A must‑follow public account sharing practical AI insights. Follow now. internet + machine learning + big data + architecture = IMBA
From a simple black‑box view where an agent receives a user request and returns an answer, the article breaks down modern AI agent designs—detailing the pure Chain‑of‑Thought reasoning loop, the ReAct reasoning‑acting cycle, tool integration, iteration tuning, and how to choose the optimal architecture for production.
The study shows that LLM agents' apparent intelligence stems more from the amount and type of context they can access than from genuine reasoning ability, as demonstrated by the ContextEval framework’s controlled experiments across multiple hyper‑parameter optimization benchmarks.
This article explains why TPU programming differs from GPU, describes the explicit HBM‑VMEM‑register data movement required on TPU, introduces the Pallas grid‑BlockSpec‑Ref model, and walks through four progressively more complex kernels—including element‑wise add, tiled dot product, fused RMSNorm with scratch memory, and a production‑grade FlashAttention implementation—showing how each kernel maps to the TPU memory hierarchy and leverages Pallas features such as input_output_aliases and PrefetchScalarGridSpec.
The article reverse‑engineers the @tencent‑weixin/openclaw‑weixin npm package to reveal the full ilink API (five POST JSON endpoints), explains hidden required fields, demonstrates a QR‑code login flow, and provides a complete 120‑line Python client that can send and receive messages reliably.
The article analyzes why standard prompt tuning (CoOp) causes catastrophic forgetting in visual‑language models, introduces the KgCoOp framework that adds a knowledge‑guided loss to regularize prompts, and shows through extensive experiments on 11 benchmarks that KgCoOp improves unseen‑class accuracy, harmonic mean, and efficiency while discussing trade‑offs and limitations.
This article explains why numeric feature engineering is essential for machine‑learning models, outlines the two main challenges of differing magnitudes and outliers, and demonstrates four scaling methods—StandardScaler, RobustScaler, PowerTransformer, and MinMaxScaler—using the California housing dataset, complete with code, visualizations, and guidance on when each method is appropriate.
This article introduces nine third‑party Python libraries—glom, boltons, beartype, result, whenever, pyinstrument, dirty‑equals, stamina, and pyfunctional—that address recurring pain points such as nested data access, missing stdlib features, runtime type safety, error handling, timezone bugs, performance profiling, testing assertions, retry logic, and data pipelines, showing concrete code examples and practical benefits.
This article provides a comprehensive guide to Claude Code’s command system, covering all slash commands, CLI flags, keyboard shortcuts, hidden features, and practical workflows, showing how to initialize projects, manage context, switch models, control costs, and automate development tasks efficiently.
The article analyzes why large language models are inherently stateless, outlines a four‑layer memory architecture for AI agents—including working, situational, semantic, and procedural memory—and explains write, retrieval, and forgetting mechanisms along with current tooling such as Mem0 and Letta.
The article analyzes how most RAG pipelines blindly feed retrieved documents to LLMs, introduces CRAG's lightweight evaluator with confidence thresholds, describes its sentence‑level decomposition, filtering, and dual‑knowledge routing, and provides a full implementation walkthrough with a real insurance query example.
