This article surveys the most influential large language models released since 2017, detailing the core innovations of Transformer, BERT, GPT series, T5, Retrieval‑Augmented Generation, and the latest LLaMA and Meta models, while highlighting their architectures, training paradigms, and impact on NLP research.
This guide explains why custom output parsers are needed for LangChain when dealing with non‑JSON or XML responses, walks through inheriting BaseOutputParser, implementing parse() and optional format instructions, and provides a complete Python example that converts a simple "Key: Value" string into a dictionary.
This tutorial demonstrates how to integrate multiple large language models from different providers into a Spring Boot application using Spring AI, configure primary, secondary, and tertiary models, and implement a fallback mechanism with Spring Retry to ensure high availability of the chatbot.
This article examines the Language Self‑Play (LSP) approach for data‑free training of large language models, detailing its challenger‑solver game formulation, advantage calculations, loss functions, self‑reward extension, experimental setup on AlpacaEval, and results that show LSP can match or surpass data‑driven baselines.
This article explains Retrieval‑Augmented Generation (RAG), an AI framework that combines external knowledge retrieval with large language models, covering its motivations, data preparation, chunking strategies, vectorization, storage, query processing, retrieval, reranking, prompt engineering, and LLM generation, plus practical optimization tips.
Ling-1T, a trillion‑parameter flagship non‑thinking model, combines 50 billion active parameters per token, 128 K context, Evo‑CoT reasoning, and FP8 mixed‑precision training to achieve state‑of‑the‑art performance on complex reasoning, code generation, and multimodal tasks while outlining its architecture, benchmarks, limitations, and future roadmap.
This article explains the definitions, technical implementations, functions, limitations, and collaborative workflow of short‑term and long‑term memory in large‑language‑model agents, detailing context windows, attention mechanisms, vector storage, retrieval strategies, and future research directions for building personalized, continuously learning AI agents.
The article analyzes the security challenges introduced by the open Model Context Protocol (MCP) ecosystem, outlines typical attack vectors such as command‑execution hijacking and indirect prompt injection, and presents MCPScan—an open‑source scanner that combines static taint analysis with LLM‑driven reasoning to detect exploitable tool chains before deployment.
The paper introduces TradingGroup, a five‑agent LLM‑based quantitative trading framework that incorporates a self‑reflection mechanism, dynamic risk management, and an automated data‑synthesis pipeline, and demonstrates superior cumulative returns, Sharpe ratios, and lower drawdowns than rule‑based, ML, RL, and existing LLM strategies on five real‑world stock datasets.
This article explains how an AI‑driven multi‑agent platform automates fault postmortem generation, enriches analysis with memory management, prompt engineering, and RAG techniques, and delivers actionable insights for SREs, developers, and non‑technical stakeholders, ultimately shifting incident handling from reactive to proactive.
This article explains how the browser‑use/qa‑use platform leverages large language models to let testers describe test cases in natural language, automatically generates browser actions, executes them, and provides detailed reports, dramatically reducing script maintenance and boosting testing efficiency.
This article reviews Kuaishou’s two‑year exploration of large‑model techniques in advertising, detailing the content‑domain estimation challenges, how multimodal and LLM approaches improve full‑domain behavior utilization and external knowledge integration, and introducing the COPE product‑content representation framework and the LEARN LLM knowledge‑transfer system.
This article explains why LLMs output plain text, introduces LangChain output parsers as the bridge to structured data, details their workflow, reviews built‑in parsers, and walks through a complete Python example that builds a prompt‑model‑parser chain to generate a JSON‑based joke.
The article explains the differences between AI Agents and AI Workflows, compares their characteristics, introduces the hybrid Agentic Workflow concept, and offers practical recommendations for building enhanced LLM applications using simple prompts or advanced frameworks.
Bilibili’s massive video platform relies on a five‑layer, storage‑compute separated big‑data architecture, handling hundreds of thousands of daily tasks, and now leverages large‑language‑model assistants to automatically diagnose and resolve frequent task failures and performance slowdowns.
This article explains why prompt templates are essential for LLM interactions, compares LangChain's PromptTemplate and ChatPromptTemplate, and walks through a complete Python example that prepares the environment, builds reusable templates, formats messages, and integrates them into an LCEL chain for chat models.
This article explains why and how to create custom LLM and chat model wrappers in LangChain by inheriting base classes, implementing core methods like _call and _generate, and demonstrates a full example with an echo‑based chat model, including streaming support and testing benefits.
This article explains how function (tool) calling lets large language models like GPT or Gemini invoke external APIs, walks through defining tools with LangChain, and demonstrates a complete Python example that fetches real‑time weather data and returns a natural‑language answer.
This article analyzes a recent paper that unifies supervised fine‑tuning (SFT) and reinforcement learning (RL) for large language models under a single gradient estimator, introduces the Unified Policy Gradient Estimator (UPGE) and the Hybrid Post‑Training (HPT) algorithm, and demonstrates their superior performance on math reasoning benchmarks.
This article explains LangChain's two core abstractions—LLM for simple text completion and ChatModel for multi‑turn conversational AI—detailing their input/output formats, practical code examples, and why ChatModel is generally preferred for modern dialogue applications.
This article outlines Kuaishou's two‑year exploration of large‑model techniques in advertising, detailing challenges of sparse cross‑domain data, the COPE unified product representation framework, and the LEARN LLM knowledge‑transfer approach that together improve ad system effectiveness.
A recent interdisciplinary study from Caltech, Cambridge and others shows that while large language models can present idealized personalities on questionnaires, their actual behavior in tasks diverges sharply, exposing a ‘personality illusion’ that challenges current AI alignment approaches.
Over the past two years, Kuaishou has leveraged multimodal large‑model techniques to overcome sparse advertising data, integrating full‑domain user behavior and external knowledge via the COPE unified product representation framework and the LEARN LLM knowledge‑transfer system, achieving measurable business gains.
Learn how to serialize LangChain components—including prompts, chains, and agents—using JSON and YAML, enabling reproducibility, collaboration, persistence, and decoupling, with step‑by‑step code examples for dumping objects to files and loading them back into executable LLM pipelines.
AlphaAgents introduces a role‑based multi‑agent framework—Fundamental, Sentiment, and Valuation agents—leveraging LLMs to analyze 10‑K reports, news, and price data, with a debate mechanism via Microsoft AutoGen; experiments on 15 tech stocks show superior cumulative returns and Sharpe ratios under risk‑neutral and risk‑averse settings compared to single‑agent baselines.
This article explains the nature, types, and causes of hallucinations in large language models used for NL2SQL, reviews both unsupervised and supervised detection methods, and introduces an efficient token‑confidence based Active Sampling Detection (ASD) approach with practical deployment examples and future research directions.
This article details the engineering practice of constructing a complete data infrastructure for large‑language‑model (LLM) applications using Alibaba Cloud SLS, covering the observability challenges of the Dify platform, the redesign of the architecture, and the resulting improvements in monitoring, diagnosis, and quality optimization.
LangChain’s callback system lets developers hook into every stage of an LLM chain— from chain start/end to token generation—using built‑in handlers like StdOutCallbackHandler or custom handlers derived from BaseCallbackHandler, with examples showing constructor‑level and request‑level attachment, plus a custom handler implementation.
This article examines the challenges of processing massive multimodal data in enterprises and presents a knowledge‑augmentation framework that leverages Retrieval‑Augmented Generation, memory‑inspired architecture, and feedback loops to enable reliable, scalable AI‑driven decision making across diverse business scenarios.
This article explains Bilibili's implementation of a large‑language‑model based intelligent assistant, detailing the platform's five‑layer architecture, the huge volume of offline and real‑time jobs, common user issues like task failures and slowdowns, and how AI can help automate troubleshooting.
This article explains why initial retrieval in Retrieval‑Augmented Generation often yields noisy results, describes how rerankers act as quality filters to improve relevance, compares the leading 2025 reranker models—including Cohere, bge‑reranker, Voyage, Jina, FlashRank, and MixedBread—and provides code snippets, evaluation metrics, and guidance for selecting the right model for specific use cases.
A lightweight, real‑time platform called Awesome AI Agents Live, built by East China Normal University students, aggregates the latest AI Agent research, offering targeted navigation, structured classification, multi‑dimensional sorting, and concise paper summaries for researchers, engineers, and students.
JD Cloud has released JoyAgent‑JDGenie, the first fully product‑grade open‑source Java multi‑agent system that bundles front‑end, back‑end, framework, engine and core agents, supports major LLMs, offers layered architecture, Docker or manual deployment, and showcases demos such as PPT generation and sales analysis.
This article explains Retrieval‑Augmented Generation (RAG), an AI framework that combines traditional information retrieval with large language models, detailing its core workflow—from knowledge preparation, chunking, and embedding to vector database storage and the question‑answering stage—while highlighting key challenges, tools, and optimization strategies.
This article explains Retrieval‑Augmented Generation (RAG), an AI framework that combines traditional information retrieval with large language models, covering its core workflow—from knowledge preparation, data cleaning, and metadata extraction to query preprocessing, vector retrieval, reranking, information integration, and final LLM generation, while also reviewing common embedding models and vector databases.
This article breaks down the GPU memory requirements of large language models during training and inference, detailing the contributions of model weights, optimizer states, activations, KV cache, and activation recomputation, and provides concrete formulas, examples, and scaling insights for models like Qwen3 and DeepSeek V3.
This article presents a complete prompt‑engineering system for AI‑driven software development, detailing four core strategies, practical code examples, and formulaic guidelines that help developers move from random trial‑and‑error to precise, reliable LLM‑assisted programming.
This article explains how LangChain’s RunnableParallel component enables true parallel execution of independent sub‑tasks, walks through concrete Python examples, compares serial versus parallel runtimes, and outlines when and why to apply this pattern for faster, more capable LLM applications.
This article examines the challenges of building production‑grade AI agents—such as context drift, knowledge leakage, and fragile state handling—and presents a disciplined architecture that combines code locks, attention anchors, and Redis‑backed state management to turn a prototype travel planner into a robust, industrial‑strength system.
This article explains how Bilibili leverages a large‑language‑model‑driven assistant to diagnose and resolve failures and slowdowns in its massive big‑data platform, detailing the platform’s five‑layer architecture, common task issues, and the need for intelligent troubleshooting tools.
This article explains how AI agents work, examines OpenManus’s comprehensive tool framework, reviews Gemini CLI’s minimalist tool scheduling and error handling, and discusses Shopify Sidekick’s scaling challenges and Just‑in‑Time instruction strategy, offering practical guidance for building robust, production‑ready agentic systems.
The article explains the concept of AI‑native applications, distinguishes them from AI‑plugin extensions, outlines their core principles such as model‑first design, data flywheel, event‑driven agents, multimodal semantics, continuous learning, and provides a seven‑step practical guide with code examples for building an AI‑native app.
This comprehensive guide walks you through the fundamentals of Dify, an open‑source LLM application platform, its key features and use cases, and provides step‑by‑step instructions for preparing the environment, installing Docker and Docker‑Compose, and deploying Dify on a CentOS 7.9 server.
An in‑depth guide compares LangChain, LangGraph, and LangSmith—the three specialized layers for LLM app development—detailing their core architectures, ideal use‑cases, and how to select the optimal combination, illustrated with a real‑world iPhone 17 refund workflow example.
This article explains how Retrieval‑Augmented Generation (RAG) combines large language models with vector databases to let non‑technical staff query massive membership data using plain language, detailing the workflow, technical architecture, optimization challenges, and real‑world impact on data‑driven decision making.
The article explains why RAGFlow needs end‑to‑end link tracing, introduces OpenTelemetry’s core concepts, shows how custom tracing utilities are implemented in Python, describes the layered architecture, provides concrete Docker and YAML configurations, and offers best‑practice guidelines for performance monitoring and fault diagnosis.
This article provides a comprehensive technical overview of RAGFlow, an industrial‑grade Retrieval‑Augmented Generation platform, detailing its architecture, core components such as DeepDoc, intelligent chunking, embedding integration, multi‑stage retrieval, and agent workflow, while comparing it with traditional RAG shortcomings.
This article examines RAGFlow's end‑to‑end pipeline for turning diverse documents into structured knowledge, detailing the TaskExecutor factory, the DeepDoc layout‑aware parser, chunking strategies, embedding and storage mechanisms, and the GraphRAG‑based knowledge‑graph extraction that together enable high‑precision retrieval and reasoning.
This guide walks you through installing the LangChain library, adding model‑specific packages, verifying the setup with a Python script, configuring API keys via environment variables or a .env file, and preparing to use OpenAI‑compatible models such as DeepSeek or Qwen.
This repository offers a comprehensive, step‑by‑step LangChain tutorial series that walks developers through installation, the LangChain Expression Language, streaming, parallel execution, callbacks, serialization, model customization, prompt templates, memory, multimodal support, and advanced tools like LangGraph and LangSmith, enabling the creation of sophisticated AI applications.
The article explains how Coze's Compass introduces a flexible evaluation system for AI agents, outlines a four‑dimensional submodule assessment (planning, tool use, self‑reflection, memory), and details specific testing criteria and challenges for web, scientific, dialogue, and programming agents.
This article explores the problem of large‑model hallucinations, explains its definitions and classifications, analyzes root causes in data, algorithms and inference, and presents detection methods and practical mitigation techniques such as RAG, decoding strategies, and model‑enhancement approaches, illustrated with real‑world 360 use cases and future research directions.
The article introduces CID-GraphRAG, a novel framework that combines intent‑driven graphs with semantic similarity search to improve multi‑turn intelligent customer service, detailing its architecture, dual‑layer retrieval mechanism, evaluation against baseline models, and future research directions.
This article walks through installing MindsDB, configuring its unified data access layer, and demonstrates how to query across relational databases, files, and vector stores while injecting AI models—including traditional ML, LLMs, and embedding models—directly into SQL for intelligent data retrieval and analysis.
This article guides front‑end developers through the evolution of AI agents—from early chatbots to modern multimodal agents—covering LLM fundamentals, prompt engineering, LangChain.js workflow creation, Retrieval‑Augmented Generation, model context protocols, and future multi‑agent technologies.
This guide explains how Alibaba Cloud AI Gateway provides traffic management, passive health checks, first‑packet timeout, and fallback mechanisms to keep large language model services highly available during traffic spikes and overload scenarios.
This guide explains why naïvely adding special tokens during supervised fine‑tuning can destabilize a large language model, and provides step‑by‑step strategies—including tokenizer updates, embedding resizing, smart initialization, and LoRA‑based PEFT—to integrate new tokens while preserving the model's original capabilities.
This article reviews the FinKario paper, which introduces an event‑augmented financial knowledge graph and a two‑stage RAG retrieval strategy that together enable real‑time knowledge updates and efficient integration of long‑form research reports, yielding a Sharpe ratio of 4.9 and outperforming baseline LLMs and institutional strategies in back‑testing.
Ant Insurance’s tech team leveraged reinforcement learning, focused data selection, and a multi‑dimensional reward system to dramatically reduce hallucinations in LLMs, achieving top‑rank performance on the HHEM leaderboard and robust improvements across instruction‑following and reasoning‑enhanced models.
This article examines Tencent's large language model deployments across diverse business scenarios, detailing core use cases such as content generation, intelligent customer service, and role‑play, and explains the underlying technologies—Supervised Fine‑Tuning, Retrieval‑Augmented Generation, and intelligent agents—that enable these applications.
The article recounts how a risk‑control algorithm engineer progressively refined a generic large language model through four stages of prompt engineering—role‑playing, business knowledge injection, deeper analysis, and a double‑hypothesis decision framework—to transform it into a precise e‑commerce fraud detection expert.
This presentation by Yao Yitong of China Telecom AI Research Institute explains why Function Call is essential for LLM deployment, outlines data‑centric and training‑centric optimization methods, discusses common pitfalls and reward‑function design for reinforcement learning, and showcases practical Agent application patterns for real‑world tasks.
This article summarizes Quincy Qu's COLING 2025 industry‑track paper that proposes a three‑stage multi‑task LLM framework for large‑scale NLP services, achieving performance comparable to single‑task models while reducing overall service cost by up to 90.9%.
This article compares LangChain, LangGraph, and LangSmith—three complementary frameworks for building LLM-powered applications—explaining their distinct architectures, use cases, and features, and also introduces related concepts such as RAG, MCP, A2A protocols, hierarchical memory systems, context engineering, and knowledge graphs to guide developers in selecting and integrating the appropriate tools.
This article details how Zhuanzhuan leverages large language models to enhance on‑site device inspection through a three‑stage pipeline—intent understanding, retrieval‑augmented generation QA, and automated knowledge‑base upkeep—highlighting technical innovations, workflow integration, and the resulting operational benefits.
This article examines the design philosophy behind the “Bash Only” category of the SWE‑bench benchmark, explaining how its minimal‑agent approach isolates LLM reasoning by restricting interactions to a plain Bash shell, making it a rigorous, reproducible test of true software‑engineering intelligence.
At the ACM CCS 2025 live paper showcase, three groundbreaking studies—Gibbon’s fast secure two‑party GBDT training, Panther’s efficient private approximate nearest‑neighbor search on a single server, and PromeFuzz’s knowledge‑driven LLM approach to fuzzing harness generation—are presented, highlighting significant performance and security advances.
This article explores how integrating ASR, TTS, and large language models into video conferencing creates an intelligent collaboration hub that boosts efficiency, enhances user experience, expands multilingual scenarios, and provides practical architecture and Python code examples for real‑time smart meetings.
This article explains how Baidu Map’s AI navigation system converts spoken queries into accurate map instructions by combining speech recognition, intent parsing, large‑language‑model reasoning, tool calling, and memory‑reflection techniques, showcasing the underlying technologies that enable instant, context‑aware responses.
This survey formalizes the shift from preference‑based reinforcement fine‑tuning to Agentic Reinforcement Learning, defines Agentic RL via MDP/POMDP abstractions, proposes a dual taxonomy of capabilities and task domains, compiles over 500 recent works, and outlines open challenges for scalable, robust AI agents.
This article examines the necessity of integrating Supervised Fine‑Tuning (SFT) with Reinforcement Learning (RL) for large language models, surveys alternating, sample‑reuse, simultaneous, and hint‑guided fusion methods, presents the underlying loss functions, and discusses practical trade‑offs such as entropy collapse and importance‑sampling corrections.
This article debunks common misconceptions about the Model Context Protocol (MCP), explains its client‑host‑server (CHS) architecture, shows how the Host drives AI decisions while Server and Client remain model‑agnostic, compares MCP with Function Calling, analyzes SDK source code, evaluates practical trade‑offs, and outlines the true engineering value and costs of using MCP in AI applications.
This article explains the architecture, core components, and step‑by‑step implementation of a tool usage module that enables AI agents to standardize, select, execute, and transform external tools, illustrated with a sales data analysis case and detailed code snippets.
The article reviews the MM‑DREX framework, which tackles the non‑stationarity of financial markets by modeling trading as a POMDP, employing a vision‑language model‑driven dynamic router to allocate four heterogeneous experts, and demonstrates superior returns, Sharpe ratios, and drawdown control across stocks, futures, and crypto compared with 15 strong baselines.
Hugging Face AI Sheets lets users employ large language models through a spreadsheet‑like interface to clean, transform, enrich, and generate datasets without writing code, offering both zero‑shot dataset creation and import‑based bulk processing, with optional self‑hosting via Docker for privacy‑sensitive workflows.
Alibaba's Qwen team unveils Qwen3-Next, a hybrid expert LLM with 800 B parameters but only 30 B active, delivering training costs under one‑tenth of comparable dense models and more than ten‑fold inference throughput for long contexts, while matching or surpassing larger models on benchmark tasks.
The article analyzes why large language models frequently generate confidently wrong answers, attributing hallucinations to statistical inevitability, data scarcity, and limited model expressiveness, and shows how RLHF exacerbates the problem by rewarding guesses, then proposes confidence‑threshold and "I don't know" strategies to mitigate it.
This article summarizes four recent AI research papers that explore zero‑shot PDE extrapolation with text‑trained LLMs, causal hidden‑state interventions for rare financial events, tabular reformulation of graph node classification, and a multimodal model for financial time‑series forecasting, detailing their methods, experiments, and key findings.
This article details how Pinterest’s senior staff engineer Dr. Luo leveraged the open‑source Ray framework to overcome LLM data‑preprocessing bottlenecks, describing the system’s architecture, key features such as map_batches, Carry‑Over Columns and Accumulators, and the dramatic performance and cost improvements achieved.
This article examines the context isolation strategies employed by Gemini CLI and Claude Code in AI agents, detailing why isolation is essential, the multi‑layer memory architecture, tool execution pipelines, concurrency controls, session management, and practical recommendations for building robust, cost‑effective agent systems.
This article clarifies why large language models, generative AI, AI agents, and autonomous AI are distinct technologies, outlining their unique missions, capabilities, and limitations to help developers choose the right approach for building intelligent systems.
This article examines the growing problem of hallucinations in large language models, outlining their causes across the model lifecycle, classifying four main hallucination types, and presenting both retrieval‑augmented generation and detection techniques—white‑box and black‑box—to reduce factual errors in critical applications.
ComRAG tackles the static‑knowledge gaps, uneven QA quality, and storage explosion of community question‑answer platforms by integrating a static documentation vector store with dual dynamic CQA stores managed via a centroid‑based memory, delivering higher accuracy, lower latency, and scalable storage for industrial retrieval‑augmented generation.
REFRAG (REpresentation For RAG) introduces a novel decoding framework that compresses, senses, and expands context using precomputed chunk embeddings, achieving up to 30.85× faster first-token generation and 16× larger context windows without sacrificing perplexity, as validated across diverse long‑context tasks.
Google researchers have built a breakthrough AI system that uses large language models and tree‑search to automatically write, rewrite, and optimize scientific computing code, delivering solutions that surpass human experts across biology, epidemiology, remote sensing, neuroscience, time‑series analysis, and computational mathematics.
This article explores how abstract syntax trees (AST) can enrich large language model (LLM) based code question‑answering by providing precise structural context, detailing LLM strengths and limits, describing AST‑LLM collaboration, RAG integration, cutting‑edge models, practical tooling, challenges, standardisation efforts, and future research avenues.
This article explains how Mem0 provides a durable memory layer for AI agents, enabling them to recall past interactions, store user preferences, and continuously improve by integrating with Milvus vector databases and graph stores, complete with step‑by‑step code examples and performance comparisons.
This article explains why most AI agents underperform due to low problem‑resolution rates, critiques traditional prompting methods, and introduces the Parlant framework with conditional rule activation, dual protection, and state‑machine architecture, followed by a complete implementation example and best‑practice guidance.
The paper introduces the Evolutionary Factor Search (EFS) framework, which uses large language models to automatically generate and evolve alpha factors, turning sparse portfolio selection into an LLM‑guided top‑m ranking task, and demonstrates superior performance on multiple Fama‑French benchmarks and real‑world market datasets.
This article defines the Model Context Protocol (MCP), explains why it is needed to unify LLM and external tool interactions, details its client‑host‑server architecture, compares its communication modes, outlines Java/Spring AI support, and discusses current adoption trends and open challenges.
This tutorial walks you through why LangGraph outperforms LangChain by showing how to build a basic chatbot, integrate tool calls, and add both short‑term and long‑term memory, complete with code examples and visualizations for creating flexible, state‑driven AI agents.
This article introduces Ant Group’s new Ray‑based distributed agent framework Ragent, outlines its background and motivation, and details the four core modules—Profile, Memory, Planning, and Action—that together enable sophisticated LLM‑driven AI agents for large‑scale applications.
By 2025, Agentic AI is shifting from massive LLMs to cost‑effective Small Language Models (SLMs), driven by their comparable performance, lower latency, and dramatically reduced inference and fine‑tuning costs, as detailed through market data, model benchmarks, migration steps, and real‑world case studies.
Alibaba introduced the Qwen3‑Max‑Preview model, a trillion‑parameter LLM that boosts multilingual understanding, complex instruction handling, and tool use while cutting hallucinations, offers competitive benchmark scores, supports 262K context, and comes with tiered token‑based pricing that may limit broader adoption.
With software systems growing ever more complex, traditional text‑based code search falls short; this article explains how abstract syntax trees (AST) provide deeper structural understanding, improve query precision, enable advanced features like control‑flow analysis and knowledge‑graph construction, and outlines a full architecture for building AI‑enhanced code question‑answering systems.
The article recounts how a risk‑control algorithm engineer progressively refined a generic large language model through four stages of prompt engineering—defining roles, dimensions, structured I/O, business rules, behavior fingerprints, and a dual‑hypothesis decision framework—to transform it into a precise e‑commerce fraud detection expert.
In this detailed case study, a risk‑control algorithm engineer explains how he progressively refined prompts for a large language model—starting from a basic role‑playing instruction, adding business‑specific exemption rules, structuring input/output, and finally implementing a dual‑hypothesis decision framework—to transform the model into a reliable e‑commerce fraud detection expert.
This article explains the core architecture of AI agents powered by large language models, detailing how planning, short‑term and long‑term memory, and tool integration work together through vector databases, retrieval‑augmented generation, and summarization to enable stateful, intelligent interactions across multiple sessions.
This article examines the context explosion problem in AI agents and compares three distinct compression approaches—Manus's never‑lose philosophy, Claude Code's aggressive 92% threshold with eight‑section summaries, and Gemini CLI's balanced 70% trigger with curated history—highlighting their trade‑offs in performance, cost, and reliability.
This article compiles the ten core questions from a Tencent algorithm interview, covering overfitting, regularization, generalization error, dropout, residual connections, attention, embeddings, BART vs BERT, instruction‑tuning data, LLM hallucination, and why GANs collapse more than diffusion models, with concise explanations and interview‑ready tips.
