Short-form development notes: Next.js, React, Python, AI and large-language-model practice, plus indie-dev and engineering lessons.
I am a senior student at Nanjing University of Aeronautics and Astronautics, majoring in AI. My interests span artificial intelligence, machine learning, with a keen emphasis on computer vision and natural language processing.
Contact
GitHub · Bilibili · Email — see the nav bar and footer links above.
TLDR: After 2023, the narrative around AI online has taken on a peculiar binary structure.
On one side, there's the AGI frenzy. Every new model release triggers a wave of posts counting down to the "extinction of human professions." OpenAI's launch events now carry the ritualistic weight of Apple's keynote back in the day, with onlookers holding their breath for those few benchmark numbers—as if the moment the curve bends toward the upper right, the AGI singularity arrives tomorrow.
On the other side, there are reflections worth paying attention to. Andrej Karpathy recently mentioned on a podcast that he now uses AI-generated articles for initial screening—not to gather information, but to find the things that "haven't been written by AI yet." Gary Marcus has repeatedly discussed in public how "the average quality of AI-generated content is converging toward the mean."...
TLDR: 1. The world has already entered an era of oversupply, where distribution and marketing may matter far more than production. Driving traffic is what truly determines survival.
2. SEO is something where experience outweighs technique. The SEO tips written in books, spanning dozens of pages—internal links, DR, authority, long-...
TLDR: This article delves into the technical differences between Flutter and React Native in terms of rendering mechanisms. It begins by outlining the similarities between the two in market presence and ecosystem, then provides a detailed analysis of their respective rendering principles. By comparing core concepts such as JSX/TSX, fiber nodes versus widgets and elements, it reveals the fundamental differences between React's reliance on Chrome for DOM operations and Flutter's self-built rendering engine. Additionally, it introduces optimization techniques to reduce the complexity of diff calculations for improved performance, and summarizes the scenarios best suited for each framework....
TLDR: This article shares the author's new insights and lessons learned from independent product development, including an evaluation of Gemini's coding capabilities, the limitations of the web ecosystem, and the challenges of reusing ecosystems from interpreted languages. The article delves into how to reuse the ecosystem of interpreted languages in applications, and proposes three solutions: running on the backend, integrating with a web frontend, and building as a desktop or mobile application....
TLDR: This article systematically elaborates on the full-chain practical strategies for independent developers, from product selection to launch, covering differentiated product selection principles (blowout strategies, 3-point product teardown), geographic arbitrage market selection, minimalist MVP development principles (core features first, code freeze), ROI-driven marketing channels (Reddit/Hacker News priority), algorithm countermeasures (golden time windows, stealth verification), trust engineering construction (visual trust verification, lowering experience barriers), passive traffic layout (Programmatic SEO/AIEO optimization), and asset-oriented operational thinking. It emphasizes that developers need to break through a technology-centric mindset and establish a resource allocation model of "70% marketing + 30% development."...
TLDR: This article introduces four common methods for independent developers to select products. The first method is to search the Chrome extension store for plugins with many negative reviews and develop an improved version. The second method is to use SEO traffic interception, targeting specific audiences by developing simplified versions of products from major companies. The third method is to observe outsourcing needs on Upwork or Fiverr and turn these needs into SaaS products. The fourth method is to check sold SaaS projects on Acquire.com to validate market demand and choose a category for development....
TLDR: This article explores whether mixed precision training (AMP) truly reduces GPU memory usage. By analyzing the storage requirements of model parameters, gradients, optimizer states, and activations under FP32 and FP16, it points out that although forward inference halves memory usage, backpropagation requires converting back to FP32 due to numerical range limitations and retaining a copy of the original FP32 model, resulting in total memory usage becoming 1.5 times the original. However, in actual training, gradients and activations are the primary overhead, and comprehensive calculations show that mixed precision can still reduce overall memory consumption (e.g., a 7B model dropping from 112G to 84G) while also improving computational speed....
TLDR: This article provides an in-depth analysis of Vercel's Fluid computing technology, which addresses the cold start problem in traditional Serverless architectures through a mechanism similar to an operating system's thread pool. Its core lies in reusing already-created Node.js instances rather than frequently creating new ones, enabling features such as real-time scaling, resource pre-allocation, and pay-per-actual-usage billing, while also supporting advanced capabilities like streaming and post-response processing. The article points out that this design pattern closely resembles the resource scheduling strategies of computer operating systems, reflecting the innovative application of underlying systems engineering principles in the cloud-native domain....
TLDR: This article provides a detailed explanation of how to implement search functionality in Next.js, covering both frontend debounce design and backend database query strategies. On the frontend, a custom hook `useDebounce` is used to delay triggering search requests based on input, preventing excessive requests. On the backend, Prisma is initially used to perform fuzzy matching queries directly against the database, with a note that this can be extended to Elasticsearch in the future to handle high-concurrency scenarios....
TLDR: This article provides a detailed introduction to the caching mechanisms supported by Next.js at different levels, including response header control at the API level, caching strategies for the Fetch function, function-level caching implemented via the Cache function, page-level ISR caching, and cross-instance KV distributed caching. It focuses on analyzing the technical implementation methods and applicable scenarios of each layer, such as setting browser/CDN cache freshness through the Cache-Control header, the force-cache forced edge caching feature, and key technical points like the react.cache higher-order function for optimizing repeated computation logic....
TLDR: This article uses the Mario game as an example to provide an accessible introduction to the foundational theories of reinforcement learning (RL), including core concepts such as policy functions, value functions, and advantage functions. It elaborates in detail on the application of Monte Carlo methods and temporal difference methods in policy optimization, and explores how the PPO algorithm addresses the issue of unstable training....
TLDR: This article explores the practical integration of the Rust language with AI-assisted programming, analyzing AI's performance at different stages of development through the creation of a Bilibili downloader project. In the early stages, AI efficiently resolved syntax and basic logic issues, but as code complexity increased, problems emerged such as insufficient context understanding, reinventing the wheel, and patch-style coding. The article evaluates Rust's learning curve, the value of AI tools, and the evolving role of software engineers, noting that current AI is better suited for handling well-defined and limited task scopes....