Search the archive

DATAFUEL

DataFuel is an API that transforms websites into structured data suitable for LLMs. It has clean Markdown or JSON output optimized for RAG systems and various AI applications, including model training and knowledge base creation.

WREN AI

Wren AI is an open-source SQL AI agent that allows users to extract data insights through natural language queries in multiple languages. It integrates with tools like Excel and Google Sheets and uses a semantic engine to understand business context and generate accurate SQL queries.

DEVIN REVIEW: IS IT A BETTER AI CODING AGENT THAN CURSOR?

Devin, a new $500/month AI coding agent accessed via Slack, is compared to Cursor, an AI code editor. Devin is able to clone repositories, generate code, and even deploy previews, but its Slack-based workflow is cumbersome and unreliable compared to Cursor's more integrated IDE approach. Cursor's real-time feedback and local control are a much better experience than Devin's long wait times and unpredictable results.

A CHATGPT CLONE, IN 3,000 BYTES OF C, BACKED BY GPT-2

This post goes through the creation of a ChatGPT clone in approximately 3,000 bytes of C code using the GPT-2 model. The program implements GPT-2's functionality, including tokenization, transformer architecture, and inference, using a custom linear algebra library.

"RULES" THAT TERMINAL PROGRAMS FOLLOW

Some unwritten "rules" of terminal program behavior include using Ctrl-C to quit non-interactive programs and Ctrl-D to exit REPLs on empty lines. Other rules include color limitations (16 colors), readline keybinding mimicry, and the use of "-" to represent stdin/stdout.

ANDROID XR: THE GEMINI ERA COMES TO HEADSETS AND GLASSES

Google, in partnership with Samsung and Qualcomm, announced Android XR, a new operating system for headsets and glasses that uses AI to create immersive experiences and integrate easily with existing Android apps and services.

THE GPU IS NOT ALWAYS FASTER

While GPUs are great for parallel applications, a vectorized CPU can outperform a GPU for specific tasks like dot products due to factors like memory bandwidth limitations.

THE OPEN SOCIAL WEB IS THE FUTURE OF THE INTERNET. HERE'S WHY I'M EXCITED

The open social web, with platforms like Mastodon and Bluesky, offers a decentralized, user-controlled alternative to traditional social media, which is exciting because of the greater level of user freedom offered.

HTTP status codes (GitHub), non-standard JSON error responses (Facebook), and deviations from the OAuth specification in parameter usage (TikTok, Strava, and Naver).

STORING TIMES FOR HUMAN EVENTS

Storing event times on websites is surprisingly complex due to user errors, international timezone variations, and historical DST changes. The common practice of storing only UTC time loses important information about the event creator's original intent, leading to inaccuracies. The recommended approach is to store both the user's intended local time and the event's location, allowing for accurate timezone conversion and easier updates when necessary.

HOW REACT COMPILER PERFORMS ON REAL CODE

The React Compiler, a tool aiming to eliminate the need for manual memoization in React apps, is tested to see if it automatically optimizes code in real-world scenarios. Through testing on both simple examples and a real-world application (approximately 15,000 lines of code) in scenarios of initial load and interaction performance, most of the improvements are in interaction speed. While React Compiler offers performance benefits in many situations, it doesn't eliminate the need for manual optimization in all cases.

DEBUNKING THE MYTH: SSR ISN'T EXPENSIVE

The common belief that Server-Side Rendering (SSR) is expensive is a misconception. Its computational overhead is minimal compared to other aspects of web requests like database queries and authentication. SSR often reduces cloud costs by consolidating server-side processes, minimizing repeated requests and their associated overhead, unlike multiple client-side requests.

USING FAKE DEADLINES WITHOUT DRIVING YOUR ENGINEERS CRAZY

Deadlines are powerful tools to drive focus and prevent scope creep, but their effectiveness depends on how they're implemented. In a healthy environment, they inspire creativity and momentum, while poor applications of them can lead to burnout and frustration.

INTRODUCING GEMINI 2.0: OUR NEW AI MODEL FOR THE AGENTIC ERA

Google has launched Gemini 2.0, its most advanced AI model yet. Designed for the "agentic era" where AI can understand the world and act on behalf of users, the multimodal model has native image and audio output, along with native tool use. Google also launched Deep Research, which lets Gemini create and carry out a research plan to create lengthy, cited research reports.

NEXT.JS 15.1

Next.js 15.1 has stable React 19 support in both routing systems and improved error debugging with enhanced source maps. New stable features include the `after()` API for post-response tasks and experimental `forbidden()` and `unauthorized()` APIs for finer-grained authentication error handling.

REACT NATIVE SHEET TRANSITIONS

React Native Sheet Transitions is a library that provides smooth, customizable, iOS-style sheet modal transitions with gesture support for React Native, Expo Go, and web apps.

THE REAL-WORLD IMPACT OF SHARED OBJECTS IN EXPO MODULES

Expo's new Shared Objects feature improves the Expo Modules API, addressing performance issues and enabling more object-oriented API design. Previously, operations like image manipulation were inefficient due to repeated disk writes. Shared Objects eliminate this by linking JavaScript and native objects, allowing data manipulation without repeated memory copies.

FAR FROM RANDOM: THREE MISTAKES FROM DART/FLUTTER'S WEAK PRNG

This report explains three vulnerabilities that stem from a weak pseudo-random number generator (PRNG) in Dart/Flutter. The vulnerabilities affected the Dart SDK, Proton Wallet, and SelfPrivacy, allowing attacks ranging from arbitrary file access to the compromise of user accounts. The weakness arose from a 32-bit truncation of the PRNG's 64-bit seed, reducing its entropy and making it easily brute-forceable.

NEW DATABASE DETAILS AI RISKS

The AI Risk Repository, developed by MIT researchers, offers a framework for monitoring and addressing AI risks, including bias, data breaches, and misuse. It categorizes over 700 risks by cause and domain.

TAILWIND CSS JUST CRACKED 1 BILLION TOTAL DOWNLOADS

TailwindCSS, an open-source CSS framework, has hit the 1 billion download mark.

STATE OF REACT NATIVE

The State of React Native 2024 survey is now open, seeking developer feedback on popular and unpopular libraries and trends within the ecosystem.

WEB ALMANAC

The 2024 Web Almanac is an annual report by the HTTP Archive that analyzes the state of the web, with topics like CSS, JavaScript, performance, accessibility, privacy, and sustainability.

FCC OPENS ENTIRE 6 GHZ BAND TO VERY LOW POWER DEVICE OPERATIONS

The FCC opened the entire 6 GHz band to very low-power devices, expanding capabilities for technologies like virtual reality and wearable devices while protecting existing services.

HOW TO ACTUALLY MIGRATE COMPLEX SYSTEMS IN INFRASTRUCTURE

Migrating complex systems requires careful planning and execution to avoid disrupting services and user experience. Approaches like hard forking, splitting teams, or relying on single individuals are discouraged. Instead, the "Strangler Fig Pattern" is recommended, where the new system gradually replaces parts of the old system without affecting customers, minimizing downtime and risk.

RUNNING DURABLE WORKFLOWS IN POSTGRES USING DBOS

DBOS, a durable workflow engine built on Postgres, uses Postgres' features to store workflow state, providing "save points" that enable recovery from crashes. DBOS functions, written in Python or TypeScript, use decorators to manage workflow steps, storing each step's state and output in the Postgres database for deterministic execution.

INTRODUCING EFFICIENT VALKEY-BASED CACHING FOR NEXT.JS

Server-side rendering in Next.js can be slow due to its CPU-intensive nature, especially in horizontally scaled environments where caching inconsistencies arise. Valkey, an in-memory data store, offers performance gains through autopipelining, automatically batching operations. It's a solution for shared caching that provides consistent results across all instances.

FROM WHERE I LEFT

After a four-year break focused on writing and family, Redis' creator is returning to the Redis ecosystem. His decision stems from a desire to contribute to the community, address its fragmentation, and help shape the future direction of Redis development. It's a part-time role.

THOUGHTS ON TAILWIND 4

Tailwind CSS v4 has great improvements like the switch to LightningCSS and native CSS cascade layers, but this author has concerns about the performance implications of CSS variables, the potential for abuse with new descendant variants, and the lack of clarity in some class names.

smoothly. Read the blog

INTRODUCING NETLIFY'S FREE PLAN

Netlify has launched a free plan that offers 100 GB bandwidth, 300 build minutes, and other generous limits for deploying web projects.

INTRODUCING LIMBO: A COMPLETE REWRITE OF SQLITE IN RUST

Turso announced Limbo, a complete rewrite of SQLite in Rust, aiming for improved memory safety, modern architecture, and potentially surpassing SQLite's reliability through deterministic simulation testing, while maintaining compatibility and offering features like asynchronous I/O and WASM support.

ELECTRIC BETA RELEASE

Electric, a Postgres sync engine enabling real-time data replication for local apps, is now in beta after a six-month rewrite. It's used in production by companies like Google and Supabase to handle high-throughput workloads with low latency.

THE GOOGLE WILLOW THING

Google announced its new 105-qubit superconducting chip, "Willow," which demonstrated a fault-tolerant surface code qubit and a new quantum supremacy experiment. It's a major achievement in experimental quantum computing, pushing the field closer to fault-tolerant quantum computation.

HOW TO THINK ABOUT TIME

Physical time (instants and durations measured in seconds) is different from civil time (datetimes and periods based on calendar systems like the Gregorian calendar), as there are more complexities and potential for bugs when dealing with civil time. It's important to use well-defined time zones (like those from the IANA database) and avoid ambiguous types like "zoned datetime.” This guide provides language-specific API recommendations for handling date and time operations well.

DEVIN IS NOW GENERALLY AVAILABLE

Devin, an AI code assistance tool that made the news a while back for controversially claiming to be the first AI software engineer, is now generally available. It's meant to handle small frontend bugs and edge cases, first-draft PRs, and targeted code refactors.

DEVELOPING A 1.7K STAR PLUGIN WITH 25K LINES OF CODE ON A PHONE (REDDIT THREAD)

The developer of markview.nvim, an open-source Markdown previewer in Neovim, developed the entire plugin (25k lines) by coding on their phone, as they don't have a laptop.

WASM-VIPS

wasm-vips is a library for browsers and Node.js, compiled into WebAssembly used for fast and memory-efficient image processing through parallel pipelines.

B+ TREE VISUALIZATION

An interactive visualization of B+ trees.

ASK HN: THOSE MAKING $500/MONTH ON SIDE PROJECTS IN 2024 (HACKER NEWS THREAD)

This Hacker News thread discusses individuals earning $500+/month from side projects in 2024.

WHAT YOU NEED TO KNOW ABOUT SQLITE

SQLite has become a viable option as the main database for Ruby on Rails apps, especially for smaller apps. SQLite does have limitations like single-writer concurrency and reliance on a single host, but its strengths are its simplicity, ease of setup, and surprisingly good performance on modern hardware.

LESSONS LEARNED FROM RUNNING REDIS AT SCALE

Mattermost scaled to 100,000 users by integrating Redis as an external cache, solving scaling bottlenecks caused by their previous in-memory LRU cache and its inefficient cache invalidation process. Optimizing Redis usage involved choices like selecting the `rueidis` library, mandating pipelining and client-side caching for reduced latency and syscall overhead, and using `MGET`/`MDELETE` for efficient bulk operations.

A LEAP IN THE EVOLUTION OF AIRTABLE'S CODEBASE: SCALING TYPESCRIPT TO THOUSANDS OF PROJECTS

Airtable's migration to TypeScript improved development, but the sheer number of projects (nearly 3,000) slowed down typechecking. To address this, Airtable restructured its codebase into smaller, more manageable projects using Bazel and custom tools to manage dependencies and identify bottlenecks. It used TypeScript 5.5's Isolated Declarations feature, along with automated type generation and improved tooling, which overall reduced typechecking time by 65%.

WHAT CAUSES 'BAD DAYS' FOR DEVELOPERS?

A Microsoft study revealed that engineering system friction, such as unreliable tools, infrastructure problems, and slow processes, is the primary cause of "bad days" for software developers. These bad days resulted in decreased productivity, heightened stress, and self-doubt.

ON GHOST ENGINEERS AND PROOF OR A LACK THEREOF

A Stanford researcher's viral thread claimed that 9.5% of engineers at major tech companies are unproductive. Critics argue that the study's underlying research, focused on predicting code review evaluations, doesn't directly support the "ghost engineer" claim and that the methodology suffers from a lack of crucial statistical measures.

DATA SCIENCE PROJECT FAILING AFTER 1,600 DAYS

A data science project scraping German news for COVID and Ukraine war data was abandoned after 1,600 days of work due to the creator's loss of interest. Sticking with a project is hard. It's helpful to have a compelling narrative to maintain motivation through the lows.

SORA IS HERE

OpenAI has released Sora, a text-to-video generation model, as a standalone product for ChatGPT Plus and Pro users.

TODOCTOR

Todoctor is a CLI tool that collects and monitors TODO/fixme style comments in JavaScript and TypeScript repositories.

RADON IDE

Radon IDE is an extension that turns VSCode or Cursor into a full featured IDE for React Native and Expo. It has features like an element inspector with component hierarchy, debugger and logger with the ability to jump-to-source, and device settings adjustments for location, language, theme, and text size.