Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
Generators in JavaScript are special functions that allow you to pause and resume execution, making them ideal for controlling when data is processed. They are especially useful for handling large datasets, asynchronous tasks, and complex iteration patterns.
Traditional management techniques for scaling startups are fundamentally flawed and detrimental to founder-led success. This post highlights Brian Chesky's experience at Airbnb, where following conventional wisdom led to disastrous results, prompting him to develop a new approach. This new approach, "founder mode," is characterized by closer engagement with the company, even at the detail level, and a willingness to break conventional norms.
Often, people think they are stupid for asking certain questions. They shouldn't mind this, because they should prioritize learning and understanding over appearances. This willingness to ask seemingly "stupid" questions leads to deeper comprehension in the long run.
The choice between becoming a manager or staying as an individual contributor depends on your career goals and strengths. Managers focus on people management and team impact but have less control over their career growth, while ICs have more autonomy and career mobility.
Sponsored
Tired of staring at logs and opaque dashboards? There's an easier way to fix API bugs and performance issues. Join Dan Mindru (Morning Maker co-host) for a live coding session to see how he shaved off 22.3s of load time [11] using Sentry's Tracing and Session Replay. Submit your
blocks.md is a tool that takes your Markdown files and turns them into forms and web pages that are customizable, accessible, and fully localizable.
QRSVG is a small JavaScript project that renders two-dimensional bitmasks (mostly assumed to be QR codes) into SVG elements with a collection of SVG paths with defined purposes.
Algora TV is an interactive livestreaming and video sharing service for developers. It's like an open-source Twitch.
Despite the Varnish cache project having over 150,000 lines of code, the average productivity of the two main developers of the project over ten years is estimated at only 5 lines per hour. This low productivity is a result of prioritizing quality and correctness over speed, as shown by extensive testing, meticulous code style, and a focus on avoiding errors. The lack of user complaints about Varnish suggests that this approach has successfully achieved its goal of producing high-quality software.
This article goes over a developer's journey in optimizing the boot time of a Raspberry Pi Zero 2 W for a solar-powered WiFi camera. The goal was to minimize energy consumption by reducing both the boot time and the power draw during boot. Various hardware and software techniques were explored, including disabling unnecessary components, adjusting kernel settings, and creating a custom Buildroot system with a stripped-down kernel. By implementing these optimizations, the boot time was reduced from over 12 seconds to less than 3.5 seconds, resulting in a fivefold reduction in energy consumption.
In 1998, the author of this essay worked on a Y2K (the Year 2000 bug) preparedness project for a global advertising conglomerate. The project involved documenting the company's technological inventory and making sure it was ready for Y2K. Eventually, they realized the project was largely a sham designed to protect the company from potential legal liabilities.
There is a bug in Apple's Calculator app where some digits in the binary display are misaligned by one pixel, likely due to a rounding error.
Traditional agile software development practices are ill-suited for AI projects, leading to frequent failures - prioritize open communication between technical teams and business stakeholders for success instead.
StatiCrypt allows you to encrypt and password protect the content of public static HTML files, to be decrypted in-browser without any back-end, and serve it over static hosting like Netlify, GitHub pages, etc.
Building LLMs from the Ground Up is a 3-hour coding workshop on building LLMs from scratch and understanding tokenization, pretraining, finetuning, and more.
Sponsored
Companies spend $4,000+ and take 36 days to fill a position, only to lose 17% of new hires in their first month. In many cases, the problem is onboarding “sink or swim” processes.
Apple and Nvidia are in talks to invest in OpenAI as part of a fundraising round that would value OpenAI at above $100 billion. It is an unusual move for Apple, as the company doesn't usually invest in startups. Nvidia has stepped up its investment activity in the past two years, putting its money into AI-related companies. OpenAI is one of the largest users of Nvidia's AI chips.
A report from China claiming to use information from Tesla's supply chain states that Tesla is planning a Model Y refresh in the first quarter of 2025. Tesla is also reportedly planning to start production of a 7-seat Model Y at Gigafactory Shanghai soon. It is expected to be an upgraded version of the US seven-seater configuration for the Asian and European markets. Most of the changes made to the Model 3 in last year's refresh will make it to the new Model Y.
The Chinese government's current five-year plan makes comprehensive research facilities for crucial fusion projects a major priority for the country's national science and technology infrastructure. It is estimated to be spending around $1.5 billion each year on fusion, almost double what the US government allocated this year for this research. The country has built itself from being a non-player 25 years ago to having world-class capabilities. Chinese scientists are aiming to build a test reactor with the goal of producing up to 1 gigawatt of fusion power in the 2030s.
Decision making in biotech is ridiculously difficult. ML systems can automate low- and medium-level decision making. These systems can generate more data and better synthesize that data into decisions. They are able to systematize human rational frameworks into quantitative, interpretable, and scalable analytic engines. Enabling model-driven decision making would require building an organization in a data centric way with an engineering culture.
Sponsored
Are slow test cycles bottlenecking your dev teams' release velocity? With QA Wolf [10], your organization can run entire test suites in minutes to ship features faster, and with confidence.
Elasticsearch and Kibana are now open source again. AGPL will be added as another license option next to Elv2 and SSPL in the coming weeks. Elastic changed its license 3 years ago to remove a lot of market confusion, but the landscape has changed so it is able to become open source again.
This developer recently crossed over $1 million on GitHub Sponsors after leaving their day job 5 years ago with no plan. In this article, they provide the strategy they followed to reach that achievement. The developer sold screencasts of them building stuff with their product. They also documented their work in detail through writing, speaking, and recordings.
SB-1047 passed the California State Assembly by a 45-11 vote. It now faces just one more procedural state Senate vote before heading to the governor's desk. The bill asks AI model creators to implement a 'kill switch' that can be activated if a model starts introducing novel threats to public safety and security. It has been criticized for focusing on risks from an imagined future AI rather than real present-day harms like deepfakes or misinformation.
The ROI on AI has been positive so far. Scaling laws may eventually collide with economics - if they hold. If scaling laws hold and there are no fundamental technology breakthroughs, then the models trained in 2026/2027 are likely to require clusters that cost well over $100 billion. At that point, the ROI on AI may begin declining.
OpenAI and Anthropic have signed deals granting the US government early access to conduct safety testing on their new AI models before they're released to the public.
Departure mono is a monospaced pixel font inspired by the constraints of early command-line and graphical user interfaces.
Meta's Llama large language models have been downloaded almost 350 million times, with approximately 20 million of those downloads taking place in the last month alone.
This post describes the process of discovering a bug that allowed anyone with a basic knowledge of SQL injection to add anyone they wanted to airport security systems.
ChatGPT now has more than 200 million weekly active users.
Chinese tech professionals are moving to Silicon Valley for opportunities they don't believe are available in China anymore.
Yes, that's not a typo. 100 million tokens in context for reasoning and programming agents. Magic Dev also announced a partnership to build two new supercomputers on Google Cloud. This comes based on the recent $320m fundraise to accelerate its product development.
OpenAI and Anthropic have agreed to allow the US government early access to their major new AI models before public release to enhance safety evaluations as part of a memorandum with the US AI Safety Institute.
California's controversial AI safety bill, SB-1047, has passed the State Assembly and is now one step away from Governor Gavin Newsom's desk. Newsom must decide by September 30 whether to sign it into law.
Mambda models are hard to accelerate with speculative decoding because of the lack of a KV cache for backtracking. This paper, from some of the original authors, showcases some novel acceleration algorithms and distillation methods.
When serving many concurrent requests to a large language model, things often get bottlenecked because we don't know how long output generation will be, so we get head of line bottlenecks. However, you can learn to rank relative lengths between requests and serve the shortest ones first, which leads to a 6.5x higher throughput for multi batch generation.
MTMamba++ is a new model architecture designed to enhance multi-task scene understanding. This approach uses a Mamba-based decoder featuring two core blocks, STM and CTM, to capture long-range dependencies and improve cross-task interactions.
🕺 WHO'S ATTENDING: Engineers & experts from DataStax, PayPal, NVIDIA, Vercel, Unstructured & more
Sponsored
Join us for the AI event of the summer in NYC! Enjoy great food, drinks, and networking while learning from the successful AI journeys of top companies. Hear about the tools used, lessons learned, and wins on their path to production.
Great open dataset from Nous Research for training models to call functions.
The Qwen team has released vision language models built on top of Qwen 2.
RAW-Adapter is a new approach that adapts sRGB pre-trained models to handle camera RAW data effectively.
Meta has released some statistics about Llama model usage. It found strong demand for its models being deployed in enterprise settings.
This study has adapted the Segment Anything Model 2 (SAM 2), originally designed for video annotation, to streamline the annotation of 3D medical images.
Anthropic has shifted to a cautious endorsement of California's AI regulation bill SB 1047 after key amendments. Changes include reduced penalties, clearer language, and increased reporting requirements. The bill aims to ensure safety in AI practices, gaining support from Anthropic yet facing opposition from OpenAI and Meta.
Flow is a Mac dictation app that lets you speak naturally and writes in your style in every application, 3x faster than a keyboard.
OpenAI announced that ChatGPT has surpassed 200 million weekly active users, doubling its user base from a year ago, and noted that 92% of Fortune 500 companies are using its products.