English | 简体中文
Welcome to my "Today I Learned" (TIL) repository! This is a personal space to document and share concise notes on various topics I encounter and learn daily. The goal is to create easily digestible snippets of information that can be quickly reviewed and understood. Think of it as a public learning journal to track my learning journey in my free time.
This will be dynamically updated as I add more TILs.
This repository is structured by topic categories. Each "TIL" entry should be a short, focused explanation of a single concept, fact, or skill I've learned. I aim for clarity and conciseness, making each entry a quick and valuable read.
- 2025W10: Mac Studio (M3 Ultra), QwQ-32B model, Manus AI Agent, and more.
- 2025W11: Gemma3, OpenAI Agent SDK, Vibe Coding, YOLOE, and more.
- 2025W12: NVIDIA GTC 2025, IPKVM, SpatialLM, Refly, and more.
- 2025W13: DeepSeek V3 (0324), Gemini 2.5 Pro, GPT-4o Image Generation, and more.
- 2025W14: Meta, OpenAI, DeepSeek next-generation model rumors, and more.
- 2025W15: Llama 4 controversy, Google DeepResearch using Gemini 2.5 Pro as a base model, SpatialLM model analysis, and more.
- Insights from John Hennessy: From the RISC Efficiency Revolution to the AI Paradigm Shift, the Next Decade of Computing: Turing Award winner John Hennessy has observed that the field of computing is undergoing a profound transformation driven by efficiency. This transformation began with the disruption of processor design concepts by RISC architecture and is now shifting more of the burden of performance enhancement and efficiency optimization to the software layer through heterogeneous computing, domain-specific architectures (DSAs), and artificial intelligence (especially large language models, LLM), against the backdrop of the slowing of Moore's Law.
- Using Gemini 2.0 Flash for High-Quality Audio Transcription and Analysis: Using Gemini 2.0 Flash Thinking Experimental 01-21 model for high-quality audio transcription and analysis on Bilibili videos.
- Latent Space Revolution: A Deep Dive into VAE Architectures and Performance Comparison of Flux.1 and Stable Diffusion: Deep analysis and performance comparison of VAE in Flux.1 and Stable Diffusion, and other related autoencoders.
- The Rise and Fall of Crypko: Analyzing the Origin, Technology, and Shutdown of an AI Anime Character Generation Platform: A comprehensive analysis of the origin, technology, and shutdown of Crypko, an AI anime character generation platform.
- YOLOE Paper Quick Read: A new efficient unified model for real-time object detection and segmentation in open-vocabulary scenarios.
- YOLOE In-depth Analysis: A deep analysis of YOLOE, including its innovative framework, performance evaluation, technological evolution, application fields, and future trends.
- Unik3D Paper Quick Read: A new general monocular depth estimation framework that can handle any camera model.
- DeepSeek-GRM Paper Quick Read: Extending inference-time computation as an effective path to improve the performance of generalist reward models.
- Rerun and Foxglove: Emerging Data Visualization Platforms for Robotics: Introducing Rerun and Foxglove, two emerging data visualization platforms for robotics, and comparing them with RViz and Unity.
- Fusion and Evolution: How VLMs and World Models are Reshaping Autonomous Driving Technology: Deep analysis of VLM-E2E, Doe-1, and DriveVLM in the context of autonomous driving, and the comparison of VLM, VLA, and world model architectures.
While this is primarily a personal learning log, constructive feedback and suggestions are welcome! If you spot any errors or have ideas for improvement, feel free to open an issue.
This project is licensed under the MIT License. See the LICENSE file for details.