Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
Google has released new AI-powered features for Android Studio. These features allow developers to chat with Gemini, receive insights about Android software, generate code suggestions, and modify code directly using prompts.
Fine-tuning can be used on LLMs for knowledge injection, teaching the model new information, and alignment. Alignment-focused objectives can be achieved via fine-tuning on small, high-quality datasets. A lot of data isn't needed to learn the style or format of output, only to learn new knowledge.
The Claude for Desktop Beta is now available on macOS and Windows.
Methods without an explicit receiver, including user-defined ones, are always instance methods of the current object, including those defined at the top level.
The Apple M1/M2 GPU driver made progress in OpenGL conformance, tessellation support using OpenCL, and its ability to run AAA games through a combination of virtualization, emulation, and driver adaptations.
Physical Intelligence has developed a general-purpose robot foundation model, π0, which can control a variety of robots and perform tasks that no prior robot learning system has successfully accomplished, such as folding laundry or assembling a cardboard box.
Sponsored
Join cloud native enthusiasts and open source innovators at KubeCon + CloudNativeCon North America, November 12 -15 in Salt Lake City!
This developer rewrote a C++ codebase to Rust and, while it was mostly successful, there were still some interoperability challenges. Using raw pointers and unsafe blocks in Rust proved to be a major source of undefined behavior that was often undetected by the compiler and required tools like Miri for validation. The Rust borrow checker and the lack of native arena allocators created challenges when managing memory resources.
Using large contexts in React applications can lead to performance problems, as changing a single value in a large context can cause unnecessary re-renders of components. There are several proposed solutions: splitting components, splitting contexts, and using state management libraries. The `use-context-selector` library, which this article dives deeper into, provides a way to optimize performance by selecting and re-rendering only the parts of a context that have changed.
Rolling your own session-based auth in JS/TS is useful for understanding core web authentication concepts. This article goes over cookie management, password hashing using Argon2, and framework-agnostic examples for sign up, sign in, and sign out functions.
Sprints are a core element of Scrum, but can be detrimental to software development. Sprints are too rigid, prioritizing processes over individual creativity and customer value. Instead, allow teams to be more flexible about timelines and give engineers more autonomy in their work.
Patrick Mackenzie's classic blog post on salary negotiation is still relevant today. However, negotiation is becoming more challenging in this market due to companies' cost-minimization strategies and the reduced leverage of candidates. Negotiation is still done best when there are multiple offers to choose from.
Google hosted a "Creator Conversation Event" for independent site owners who had been shadowbanned by Google's algorithm updates. At the meeting, Google representatives offered reassurances while subtly deflecting and avoiding addressing the core issue of shadowbanning and the negative impact on these creators' livelihoods.
The web got updates in October like synchronous iterator helpers, text fragments, and cookies with partitioned storage in Firefox 131. Beta releases of Firefox and Chrome have WebCodecs API support, worker navigator permissions, and CSS highlight inheritance (Chrome-only).
Modus is an open-source serverless framework that allows developers to build APIs powered by WebAssembly. It simplifies the integration of AI models, data, and business logic by using a sandboxed execution environment. Modus supports multiple programming languages, like AssemblyScript and Go.
Wazuh is an open-source security platform that offers endpoint and cloud workload protection, XDR, and SIEM capabilities.
A large-scale scheme involving malicious Chrome extensions disguised as legitimate tools was discovered. Eleven extensions, seemingly developed by a mysterious individual with no online presence, have been found to contain hidden code for affiliate fraud and user tracking.
LLM-as-a-judge systems have LLMs evaluate LLM responses in order to make LLM applications better. This article outlines seven steps to building a great LLM-as-a-judge system: find a “Principal Domain Expert”, create a diverse dataset, have the expert review and critique the data, fix errors, build the LLM judge, perform error analysis, and create specialized judges if needed.
Chain-of-thought (CoT) prompting LLMs reduces performance in cases where thinking also negatively affects human performance. Experiments across three task categories where thinking harms human performance (implicit statistical learning, visual recognition, and classification with exceptions) consistently demonstrate that CoT reduces model accuracy. This suggests that model reasoning, like human deliberation, can lead to detrimental outcomes when applied to certain tasks.
The State of Frontend 2024 report shows the rise of AI, the continued dominance of React, and the growing importance of developer experience and accessibility.
The ability to easily change code is the true definition of quality in software development as it means a non-committance to irreversible decisions.
Apple has announced the new MacBook Pro, which features the powerful M4 family of chips, Thunderbolt 5 connectivity, an advanced camera, and Apple Intelligence.
Dropbox CEO Drew Houston announced a 20% reduction in the company's global workforce due to softening demand, an overly complex organizational structure, and underperformance in some areas.
Ezcrypt is a secure and portable file encryption tool that uses strong cryptography.
Cloudflare migrated its DNS database to a new cluster. The process involved separating DNS record data from zone data, creating a new internal DNS Records gRPC API, and building a custom Change Data Capture and Transfer Service (CDCTS). The CDCTS enabled near real-time data transfer. Afterwards, Cloudflare saw an improvement in its DNS Records API performance.
Netflix optimized its logged-out homepage for improved Time-to-Interactive by switching from React to vanilla JavaScript and prefetching resources for subsequent pages. This led to a 50% reduction in Time-to-Interactive on the homepage and a 30% reduction for future page navigations.
Server-Side Rendering (SSR) is where the server generates HTML content which is then sent to the browser for immediate display, leading to faster initial load times and better SEO. This article outlines two approaches to SSR in Vue: manual rendering, which requires setting up a Node.js server, and using pre-built solutions like Nuxt, Quasar, or Vite-plugin-ssr/Vike.
This developer became more visible by consistently sharing their knowledge about web development. They started by sharing their learnings on various platforms like blogs, Twitter, and Stack Overflow, focusing on making complex concepts accessible and relatable.
Successful one-on-ones are not about rigid guidelines but about three core principles: purpose, curiosity, and individualization. Having a clear purpose for the meeting, beyond simply fulfilling a company requirement, is necessary.
Microservices hype is a rehash of older architectural concepts like modules and distributed systems. The purported benefits of microservices, such as scalability and independent development, were already achieved by these older techniques.
Sponsored
Build a cross-platform full-stack React Native application using a single codebase with scalable, real-time backend support, authentication, and user management.
ml.js is a JavaScript library containing a collection of machine learning tools for use in web browsers. It has a set of algorithms for both supervised and unsupervised learning covering tasks like clustering, classification, regression, and optimization.
GitHub is expanding its AI options for GitHub Copilot by partnering with Google and Anthropic. OpenAI models remain the default, but users can switch to the Claude or Gemini models if they want. GitHub is also introducing Spark, a new AI project that enables users to create simple applications using natural language prompts.
Cursor, an AI code editor, has a helpful tab completion feature that suggests code completions and navigates to the next recommended action with impressive speed and accuracy. Other Cursor features, such as inline editing, chat sidebar, and Composer, are useful for code refactoring, generating code, and translating between different programming languages.
OpenAI is collaborating with Broadcom and TSMC to create its first in-house AI chip, focusing on inference. This move comes after OpenAI abandoned plans to build its own chip manufacturing network due to the high costs and long lead times.
React hooks, especially useEffect, rely on the concept of object stability, which is not clearly documented, making it difficult to determine which return values of hooks are stable and which are not.
This free course teaches React developers how to use Vim motions to navigate, edit, and refactor code more efficiently.
Web components are a great choice for building leaf nodes, design systems, and progressively enhancing regular HTML, but they come with challenges like Shadow DOM complexity and limitations in server-side rendering.
A lesser-known feature in JavaScript, using sibling parameters as default values in functions, is useful for code testability.
There are various ways to integrate Rust into non-Rust (i.e. Node.js) servers to improve performance. This article presents four tiers of Rust integration: Tier 1 uses a Rust CLI tool, Tier 2 uses a Rust Wasm module, Tier 3 uses a Rust native function, and Tier 4 involves a complete Rust rewrite of a server. Using a Rust native function (Tier 3) offers the best balance of performance gains and ease of implementation, while rewriting an entire server in Rust (Tier 4) achieves the highest performance (but is often impractical).
HTML forms have powerful validation mechanisms, but they aren't used much because the setCustomValidity API is cumbersome to use. A better way to make it easier to use is to use a declarative custom-validity attribute, which lets developers define input validations within their components.
Angular's approach to partial hydration ("incremental hydration”) involves deferring the hydration of certain sections, typically those not immediately visible or interactive, until needed. Angular uses deferrable views, event replay, and defer blocks to achieve this.
It's better to have a mindset of writing code that is easy to delete. To do this, you can start by copying and pasting code to avoid unnecessary dependencies, then progressively layer and modularize code to isolate complex or changeable parts. The end goal is to create code that can easily adapt to change and iteration.
What you work on is more important than how hard you work. It's better to focus on high-impact projects that align with both organizational goals and career advancement and say no to low-priority work by clearly communicating your reasoning.
This author reflects on the shift from optimistic to more realistic views of the web's potential. He critiques techno-optimism, highlighting the web's vulnerabilities to disinformation and manipulation and the concentration of power in a few tech giants. The author remains hopeful, drawing inspiration from the web's decentralized origins and the potential for new technologies like the Fediverse and Interledger to create a more equitable and open internet. Ultimately, he acknowledges the possibility that his optimism may prove naive, but emphasizes the importance of learning from past mistakes and building a better web for the future.
Sponsored
Build a cross-platform full-stack React Native application using a single codebase with scalable, real-time backend support, authentication, and user management.
Trench is an open-source, real-time event tracking system built on Apache Kafka and Clickhouse. It offers GDPR-compliant analytics and scalable deployment options for self-hosting or a fully managed cloud solution.
Radix Vue is an open-source, accessibility-focused UI component library for Vue.js that ports the Radix UI primitives.
Outlines is a library for structured text generation. It has features like multiple choices, type constraints, regex-structured generation, efficient JSON generation, and the ability to use context-free grammars to guide generation.