Ideas Worth Exploring: 2025-03-14
- Charles Ray
- Mar 14
- 4 min read
HTTP Toolkit: HTTP/3 is everywhere but nowhere
HTTP/3 and QUIC (the protocol beneath it) have been developed since 2013 and are now standardized, supported in 95% of users' browsers, already used in 32% of HTTP requests to Cloudflare, and supported by 35% of websites through alt-svc or DNS. Despite this wide adoption, neither QUIC nor HTTP/3 are included in the standard libraries of major languages like Node.js, Go, Rust, Python, or Ruby.
The benefits of HTTP/3 include reduced latency, header compression, bidirectional streaming, advanced prioritization support, increased resilience to unreliable networks, zero-round-trip connection initialization, and other improvements. However, most developers cannot easily use HTTP/3 end-to-end today due to the lack of support in open-source implementations and libraries.
This has created a two-tier web where major browsers and specific mobile app traffic communicate with a small set of big servers using HTTP/3, while everyone else uses other implementations. The long-tail web is bigger than the hyperscale web but is fragmented into different implementations and lacks the concentrated resources and central direction found in the hyperscale world.
OpenSSL's approach to QUIC has created a significant schism in the ecosystem, making it difficult for many projects to drop support for OpenSSL and adopt new implementations. This has resulted in a disparity between the benefits of HTTP/3 being enjoyed by only a small segment of the web, while others are left out.
The text suggests that this situation could have serious consequences if not addressed, including increased disadvantage for the long-tail web, widening of the split between the two tiers, and potential exclusion of non-browser clients from the modern web. The author expresses hope that a resolution will be found to improve access to HTTP/3 for all developers, but notes that for now, implementing HTTP/3 end-to-end in applications may require some difficulty due to the lack of support in open-source implementations and libraries.
GitHub Repo: Flower: A Friendly Federated AI Framework
Flower (flwr) is a framework for building federated AI systems. The design of Flower is based on a few guiding principles:
Customizable: Federated learning systems vary wildly from one use case to another. Flower allows for a wide range of different configurations depending on the needs of each individual use case.
Extendable: Flower originated from a research project at the University of Oxford, so it was built with AI research in mind. Many components can be extended and overridden to build new state-of-the-art systems.
Framework-agnostic: Different machine learning frameworks have different strengths. Flower can be used with any machine learning framework, for example, PyTorch, TensorFlow, Hugging Face Transformers, PyTorch Lightning, scikit-learn, JAX, TFLite, MONAI, fastai, MLX, XGBoost, LeRobot for federated robots, Pandas for federated analytics, or even raw NumPy for users who enjoy computing gradients by hand.
Understandable: Flower is written with maintainability in mind. The community is encouraged to both read and contribute to the codebase.
GitHub Repo: Carbon Design System
Carbon is an open-source design system built by IBM. With the IBM Design Language as its foundation, the system consists of working code, design tools and resources, human interface guidelines, and a vibrant community of contributors.
The library provides front-end developers & engineers a collection of reusable Angular components to build websites and user interfaces. Adopting the library enables developers to use consistent markup, styles, and behavior in prototype and production work.
Ideas: Can Good Design Be Mathematically Calculated?
The article discusses the ideas around whether good design can be mathematically calculated by exploring the intersection of art and science, focusing on mathematical principles like the Golden Ratio, symmetry, geometry, and grids in design.
It provides the perspective that while mathematics provides a valuable framework for designing visually pleasing compositions, it is not the sole determinant of effective designs as human elements such as intuition, creativity, empathy, storytelling, problem-solving, and meaningful experiences are essential to good design.
The article also highlights the limitations of a purely mathematical approach and emphasizes the importance of finding a balance between science and art in design. As technology evolves, new tools such as generative design algorithms and AI can assist designers in making informed decisions but should not replace the importance of a human touch in creating effective designs. Ultimately, good design requires finding a balance between science and art, logic, and intuition to create visually appealing and functional designs that meet user needs and communicate brand messages effectively.
GitHub Repo: Audio Flamingo 2
Audio Flamingo 2 achieves the state-of-the-art performance across over 20 benchmarks, with only a 3B parameter small language model. It is improved from our previous Audio Flamingo.
Audio Flamingo 2 outperforms prior SOTA models including GAMA, Audio Flamingo, Qwen-Audio, Qwen2-Audio, LTU, LTU-AS, SALMONN, AudioGPT, Gemini Flash v2 and Gemini Pro v1.5 on a number of understanding and reasoning benchmarks.
See the research article here: Audio Flamingo 2: An Audio-Language Model with Long-Audio Understanding and Expert Reasoning Abilities
Kommentare