LearnByDoing With Steven

-

قناة - البرامج: ‫٤‬

تعليم

تعليم

يتم التحديث أسبوعيًا
تعليم

تعليم

يتم التحديث أسبوعيًا
تعليم

تعليم

يتم التحديث أسبوعيًا
تعليم

تعليم

يتم التحديث أسبوعيًا

١٩ مارس

Agentic Storage: Solving AI State and Persistence with MCP

Standard AI agents currently operate like volatile memory, meaning they lose all progress once a session concludes. While technologies like Retrieval Augmented Generation allow agents to read existing data, they lack a dedicated method for saving original work or code. Agentic storage addresses this limitation by providing a persistent layer where autonomous systems can permanently store and manage outputs. To simplify this interaction, the Model Context Protocol (MCP) acts as a universal translator, allowing agents to communicate with various storage types through a standardized interface. Because giving AI write access involves risks, this architecture includes safety features like immutable versioning, sandboxing, and intent validation to prevent data loss or unauthorized actions. Ultimately, this evolution transforms AI from a temporary chatbot into a functional worker with a reliable, long-term memory.

١٩ مارس

-

٤٠ د
قبل ٣ أيام

Building a Personal Knowledge Management (PKM) system

Stop just reading—start compounding your knowledge. Here is how LLMs are transforming the "Second Brain." Building a Personal Knowledge Management (PKM) system is no longer about manual filing. By leveraging frontier LLMs, we can now automate the ingestion of articles, papers, and datasets into a structured Markdown wiki. Using Obsidian as the frontend, research becomes a continuous process of autonomous indexing, summarization, and connection-finding. Key Workflow Insights: Data Ingest: Automated indexing of raw documents and images.AI-Managed Wiki: Incremental compilation with zero manual intervention.Complex Queries: Autonomous research agents that query hundreds of articles.Visual Output: Auto-generated slide presentations (Marp) and data visualizations.As repositories grow, the future lies in synthetic data and fine-tuning to internalize this knowledge directly into model weights. #LearnByDoingWithSteven #SecondBrain #Obsidian #PersonalKnowledgeManagement #AI #LLM #Productivity #KnowledgeBase #DataScience #AIAgents All my links: https://linktr.ee/learnbydoingwithsteven

قبل ٣ أيام

-

٤ د
قبل ٢١ ساعة

Demystifying the Foundations of Neural Networks with MIT

Demystifying the Foundations of Neural Networks with MIT The revolution in artificial intelligence over the past decade has been nothing short of extraordinary. From solving insurmountable scientific challenges to hyper-realistic content generation, deep learning has moved from a computation-expensive luxury to a highly commoditized technology accessible to everyone. MIT 6.S191 (Lecture One) provides a foundational systematic review of the core intelligence hierarchy. By defining intelligence as information processing for decision-making, it clarifies the distinct roles of AI, Machine Learning, and Deep Learning. The growth we see today is driven by three pillars: massive datasets, efficient GPUs, and mature open-source software. Key takeaways include: The Layered Hierarchy: AI (Science) -> ML (Learning from Data) -> DL (Neural Networks).The Perceptron: The fundamental building block using dot products and non-linear activation.Training Mechanics: Using Loss Functions and Backpropagation for weight optimization.Generalization: Combatting overfitting with techniques like Dropout and early stopping.All my links: ⁠https://linktr.ee/learnbydoingwithsteven⁠ #DeepLearning #MIT #ArtificialIntelligence #MachineLearning #NeuralNetworks #LearnByDoingWithSteven #StevenDataTalk #AIResearch #ComputerScience #TechInnovation

قبل ٢١ ساعة

-

٧ د
قبل يومين

Summary of public notes regarding Claude Code source code

The veil has been lifted on Anthropic's Claude Code. Here is what we found in the 512k lines of leaked source code. A recent configuration error led to the exposure of nearly 1,900 TypeScript files, revealing the inner workings of one of the most powerful coding agents. From hidden experimental features like KAIROS and BUDDY to the sophisticated coordination of over 40 internal tools, the architecture is a masterclass in modern agentic design. Key technical takeaways: Runtime: Built on Bun with a React-based terminal interface.Scaling: A massive 30,000-token initial payload for high-precision tool execution.Privacy: An "Undercover Mode" that automatically scrubs AI metadata for employees.Caching: Intelligent static/dynamic prompt splitting to optimize cost and speed.This leak provides a rare glimpse into the future of LLM-driven engineering tools. While the caching bug discovered by the community has already been patched, the implications for agentic security and telemetry are profound. #LearnByDoingWithSteven #ClaudeCode #Anthropic #AIArchitecture #SoftwareEngineering #AgenticAI #TechNews #LLM #Coding #OpenSource All my links: https://linktr.ee/learnbydoingwithsteven

قبل يومين

-

٥ د
٤ أبريل

LLM Teams as Distributed Systems

This research explores the design of Large Language Model (LLM) teams by applying the formal principles of distributed computing. The authors argue that multi-agent systems face the same challenges as networked computers, including communication overhead, consistency conflicts, and task synchronization. By using Amdahl’s Law, the study demonstrates that a team's efficiency is strictly limited by how much of a task can be performed in parallel. Experimental results show that while decentralized teams are more resilient to individual "straggler" agents, they often suffer from higher token costs and coordination errors. Ultimately, the paper provides a principled framework to move beyond trial-and-error when building robust and cost-effective AI collectives.linktree: https://linktr.ee/learnbydoingwithstevenarvix: https://arxiv.org/pdf/2603.12229

٤ أبريل

-

٧ د
١٨ مارس

Connecting Agents and Data: A Guide to A2A and MCP

The provided text outlines two primary frameworks, A2A and MCP, designed to solve the communication challenges inherent in isolated AI agents. A2A (Agent-to-Agent) is an open protocol that facilitates direct interaction between different AI entities, allowing them to exchange tasks and status updates regardless of their underlying vendors. In contrast, the Model Context Protocol (MCP) provides a standardized way for individual agents to access external data sources like databases, file systems, and code repositories. While A2A focuses on the collaborative relationship between multiple agents, MCP serves as a bridge between an agent and its necessary tools or information. Both protocols utilize JSON RPC to ensure structured data exchange, but they serve different roles in the AI ecosystem. Ultimately, the source illustrates how these technologies complement each other to create a fully integrated environment where agents can both work together and interact with existing digital infrastructure.

١٨ مارس

-

٦ د
٢٤ مارس

Benchmarking Open Source VLM and OCR Model Performance

These sources explore the evolving landscape of open-source Vision-Language Models (VLMs) and their specialized application in Optical Character Recognition (OCR). The first source introduces DeepSeek-OCR, a model utilizing a unique DeepEncoder architecture to compress high-resolution visual data into a minimal number of tokens while maintaining high text extraction accuracy. It highlights the potential for optical compression to solve computational challenges associated with processing long document contexts in large language models. The second source provides a comparative benchmark of prominent open-source models, such as Qwen 2.5 VL and Gemma-3, evaluating their ability to perform structured JSON extraction from documents. This report identifies Qwen 2.5 VL as a top performer, matching the accuracy of leading closed-source models like GPT-4o. Together, the texts demonstrate a shift toward end-to-end neural architectures that replace traditional, multi-step OCR pipelines with more efficient and integrated visual-textual processing.all my links: linktree learn by doing with steven数能生智 https://linktr.ee/learnbydoingwithsteven

٢٤ مارس

-

٦ د
١٩ فبراير

Comparative Analysis of Large Model Inference Optimization Frameworks

This report provides a comparative analysis of specialized large language model (LLM) inference frameworks designed to overcome hardware limitations and high computational costs. It distinguishes between high-throughput server solutions like vLLM, which uses PagedAttention to eliminate memory fragmentation, and SGLang, which optimizes complex, multi-turn interactions through RadixAttention and structured generation. For local deployment, the text evaluates Ollama and LM Studio, highlighting how they leverage llama.cpp and the GGUF format to run models on consumer-grade hardware. The study further explores critical performance-enhancing technologies such as quantization, speculative decoding, and continuous batching. Ultimately, the sources serve as a guide for selecting the right infrastructure based on specific needs, ranging from cloud-scale API services to private local assistants.

١٩ فبراير

-

٨ د
٣ أبريل

Large Language Model-Enhanced Relational Operators Taxonomy and Benchmark Analysis

大语言模型增强的关系算子体系与基准测试分析 Large Language Model-Enhanced Relational Operators Taxonomy and Benchmark Analysis 随着大语言模型的发展，诸多研究通过类似算子的组件将大语言模型集成到关系型数据处理任务中，例如带有语义谓词的过滤器、知识增强的表格插补、数据实体匹配以及复杂的语义查询处理。这些组件在调用大语言模型的同时保留了关系型的输入和输出接口，因此被称为大语言模型增强的关系算子。然而，现有的此类算子存在定义碎片化、实现策略各异以及缺乏全面评估基准等问题。 https://arxiv.org/pdf/2603.02537 https://linktr.ee/learnbydoingwithsteven #LargeLanguageModels #DatabaseManagement #DataScience #RelationalDatabases #ArtificialIntelligence #TechPodcast #StevenDataTalk #数能生智 #数据漫谈

٣ أبريل

-

٨ د
١٤ مارس

RAG vs. Long Context: The Future of Model Architecture

RAG vs. Long Context: The Future of Model ArchitectureLarge language models are inherently limited by training cut-off dates and a lack of access to private data, necessitating methods to inject external information. The provided text compares two primary strategies for overcoming these hurdles: Retrieval Augmented Generation (RAG) and Long Context windows. RAG acts as an engineering-heavy approach that uses vector databases to find specific snippets of information, which is ideal for managing infinite data sets while reducing computational costs. Conversely, the Long Context method offers a simpler "no stack" architecture by feeding entire documents directly into the model, allowing for superior global reasoning and avoiding the risk of missing data during retrieval. Ultimately, the choice between these methods depends on whether a user requires the efficiency of targeted search or the comprehensive oversight provided by a massive context window.https://linktr.ee/learnbydoingwithsteven

١٤ مارس

-

٨ د

Learn By Doing With Steven 数能生智

موقع الويب