Provides introductions and tutorials on Bilibili-related technologies.
SABER introduces a reinforcement‑learning framework that lets large language models dynamically switch among four token‑budgeted reasoning modes, dramatically cutting inference length while preserving or improving accuracy across math, code, and logic tasks.
This article details how we built an AI‑tool ecosystem by evolving a basic online JS cloud‑function platform into a unified, reusable capability layer that integrates with Flowise, LangChain StructuredTool, and the Model Context Protocol (MCP) to provide secure, cross‑platform tool calls for agents.
This article details the design and deployment of a multi‑layer LLM system that automatically reads massive creator group chats, extracts structured insights, mitigates hallucinations with dual‑model verification, uses few‑shot prompting for stable output, and delivers real‑time risk alerts and operational reports.
This article explains what an AI Agent is, distinguishes it from traditional workflows, and demonstrates how to build and customize agents using the Go‑based Blades framework, covering core concepts, code examples, five workflow patterns, best‑practice guidelines, and reference resources.
This article analyzes the current bottlenecks of Bilibili's private messaging service, explains the technical challenges of massive data volume and traffic spikes, and presents a comprehensive multi‑layer architecture upgrade—including cache strategies, BFF refactoring, database sharding, and consistency mechanisms—to ensure scalability and reliability.
This article explores the evolution of security alert automation from manual verification to SOAR and AI-driven solutions, detailing MCP-based AI agents, integration with various security tools, practical case studies of honey‑pot, HIDS, and EDR alert tracing, and the resulting efficiency gains and future outlook.
RIVAL (Reinforcement Learning with Iterative and Adversarial Optimization) introduces an adversarial game between a reward model and a translation LLM, combining qualitative preference rewards with quantitative metrics like BLEU, to overcome distribution shift in RLHF and achieve superior performance on conversational subtitle and WMT translation tasks.
Bilibili’s game algorithm team built a four‑layer, LLM‑based translation platform that automates terminology extraction, retrieval‑augmented generation, and quality assessment, dramatically reducing localization cycles by over 85% and costs by up to 80% while supporting ten languages and ensuring consistent, culturally‑accurate game text.
This article details how Bilibili’s multimedia lab leveraged a multimodal training pipeline combining data‑compressed SFT and the GRPO reinforcement‑learning algorithm to achieve a 13.5% metric boost and secure second place in the ICCV MIPI Detailed Image Quality Assessment competition.
VibeCut introduces a novel Orchestrator‑Executor dual‑agent framework for WebCut, leveraging large language models, shared structured context, and modular tool integration to automate complex video editing tasks, demonstrating improved efficiency, transparency, and adaptability across diverse scenarios while addressing challenges of multi‑agent coordination.
