Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
A web application has server-free architecture when it doesn’t have any backend servers, containers, or virtual machines. This can be implemented using SQLite compiled to WebAssembly for database operations, running backend code in a Web Worker, and using a Service Worker to intercept requests and communicate with the Web Worker. A server-free architecture offloads work from the backend and allows for offline access to web applications.
A function within Sourcegraph's Code AI tool, Cody, was recently optimized to significantly increase its throughput and reduce memory usage. The function calculates the dot product of two vectors. The optimization journey involved various techniques like loop unrolling, bounds-checking elimination, quantization, and SIMD (Single Instruction Multiple Data) instructions.
Best practices for migrating monolithic applications to microservices in containerized Kubernetes environments include understanding the monolith's dependencies, taking an incremental approach, loosely coupling microservices, and implementing end-to-end observability. DevOps teams are commonly under pressure to make this migration to optimize the software delivery life cycle, reduce fragility, and create shorter release cycles.
Managing up is the skill of influencing people who are more senior to you, such as your direct manager. Failing to deliver on promises and deadlines can lead to a lack of trust from your manager and potentially getting fired. To manage up effectively, it’s important to set realistic expectations, prioritize important promises, and communicate openly with peers.
Companies should hire more junior engineers as they bring fresh energy, motivation, and openness to new technologies to a team. Juniors are necessary for long-term excellence in software engineering. They should be paid fairly based on their skills and not just years of experience, or else companies may lose their best engineers over time.
TypeSpec is a language designed for describing cloud service APIs and generating various assets, such as API description languages, client and service code, and documentation. It offers extensible core language features that can describe API structures commonly found in REST, OpenAPI, GraphQL, gRPC, and other protocols. TypeSpec enables the creation of reusable patterns for API design and allows for the detection of known anti-patterns, helping API designers adhere to best practices.
Atlas is a browser-based tool that allows users to work with geospatial data and share their results. It enables users to build interactive maps, analyze data in the context of location, visualize trends, facilitate teamwork and information sharing among GIS professionals, and present plans and proposals in an easily understandable format.
Difficulty in machine learning (ML) is not primarily due to mathematics, but rather the need to build an intuition for selecting the right tools and debugging. Debugging in ML is exponentially harder than in standard software engineering due to multiple dimensions of potential issues involving algorithms, implementation, data, and model correctness. The long debugging cycles in ML, which can be days or even weeks, make parallel experimentation necessary.
A well-structured onboarding process not only grants access to new hires, but also integrates new employees socially and culturally. A bottleneck in the onboarding process can be identified by delays in accessing tools and systems, the inability of new developers to make production deployments quickly, newcomers feeling orphaned, excessive focus on individual work, resistance to change, slow completion of seemingly simple tasks, high turnover of newcomers, and documentation that can't answer new hires' questions. An optimized onboarding process maximizes the productivity of new employees.
Bufferbloat in computer networks can cause delays and glitches in internet traffic, particularly during activities like video calls. Bufferbloat occurs when network buffers fill up with data packets, causing increased latency. Active Queue Management (AQM) techniques like CAKE can be used to dynamically adjust and manage network queues, resulting in more stable and responsive network performance.
Full offline support can be supported through architectural decisions, including multi-storage loading for speed and local caching, grouped updates through an operations queue to optimize real-time interactions, and a UI status system to inform users about offline functionality.
An introduction to core concepts in cryptography using code examples with Node’s crypto module.
The gaming market had great public releases in 2023, but it also had lots of layoffs and scrapped projects behind the scenes.
Analysts are estimating that Apple Vision Pro sales were between 160,000 to 180,000 units over the first three days. The estimate is over 40% of the 400,000 units Apple is expecting to sell in the first year. The estimate is based on pre-order inventory and shipping times. Apple likely didn't have enough stock on hand to meet initial demand, but demand and supply now appear to be balanced.
MrBeast posted a video to X last week with the explicit goal of seeing how much ad revenue a video on the platform would make. The video made $260,000 with more than 150 million views. While the video performed well, MrBeast noted that advertisers bought ads to run specifically on his video, earning him a higher rate. There were also multiple reports that X was quietly boosting the video to viewers, artificially inflating the view count.
Disney's HoloTile is a system composed of hundreds of small round tiles, each serving as a kind of mini omnidirectional treadmill. The tiles work together to prevent users from walking off the HoloTile. Users can walk on the HoloTile in any direction they want and it will automatically do whatever it needs to keep them on the floor. It supports multiple people walking independently. The technology will likely become part of a Disney Parks VR experience. Video of the HoloTile in action is available in the article.
The International Thermonuclear Experimental Reactor (ITER) is a fusion reactor that may one day provide the Earth with unlimited energy. The project started in 2005. Fusion reactions occur when two small particles collide, forming a heavier atom and producing a significant amount of energy without creating radioactive waste. If successful, ITER will be arguably the most complex machine ever designed.
This repository contains tips and resources to help people prepare for behavioral interviews. It covers general tips, questions that you may be asked, and questions you can ask the interviewer. Each interview question comes with prepared answers.
The Machine Learning Engineering Open Book is an open collection of methodologies to help with the successful training of large language models and multi-modal models. The material is suitable for LLM/VLM training engineers and operators. It contains lots of scripts and copy-n-paste commands to enable readers to quickly address their needs. The book covers key hardware components, performance, operation, development, and much more.
While GPT-4 is proficient in many tasks, it is also imperfect in odd ways. For example, it can speak all languages, but barely do math. As AI becomes more competent, the technology will have an increasingly large impact on the world. This post lists several ways AI could impact the world and asks some important questions about how these impacts will change the world. The areas affected by AI include markets, real estate, energy, data centers, nations, inflation, and geopolitics.
Dopamine does not generate pleasurable feelings. It's the other way around - pleasurable feelings generate dopamine. The neural circuits that lead us to like things are separate from the circuits that generate wanting. It is easy to confuse 'liking' and 'wanting'. The brain is a prediction machine for rewards, and dopamine represents errors in predicting rewards. When bets by the brain are resolved, dopamine serves as a signal for whether the outcome was better or worse than expected and how much better or worse the outcome was than expected. The higher the surprise, the higher the reward prediction error - resulting in higher activity of dopamine neurons.
The X Lab, Alphabet's 'moonshot' experimental group, is part of the company's 'Other Bets' group, which burns through a billion dollars every quarter.
This post presents a system for generating 'random enough' numbers in your head within half a minute.
AST SpaceMobile, which already had Vodafone as an investor, announced Google and AT&T's funding along with a $100 million public offering of its stock on the same day.
This article presents 12 CSS properties to start incorporating to reduce technical debt, remove JavaScript, and score easy wins for user experience.
The Apple Vision Pro could change how media is consumed in the future - its success depends on its apps, but there aren't many developers who are particularly motivated to build experiences for the platform.
Stability AI's latest release, Stable LM 2 1.6B, is a compact yet powerful language model that supports seven languages. It is designed to outperform other models under 2 billion parameters, including its own previous 3B model. With its smaller size, it aims to lower barriers for developers, offering various versions including a unique "half-cooked" model for more customization flexibility.
Tesla has started rolling out its Full Self-Driving Beta v12 update, which shifts vehicle control from explicit C++ code to an AI-powered single neural network. The release marks a significant move towards fulfilling the company's self-driving ambitions, although the software is still labeled as beta. Skepticism will remain until tangible improvements in autonomous driving capabilities are observed as the software is rolled out to cautious select beta testers.
LoRAs are Low Rank Adapters that allow you to fine-tune only a small amount of parameters in a language model. They can dramatically improve and alter the performance of these models.
Grouping in 3D is a challenging and ambiguous task because you don’t know what granularity you’ll need for the grouping operation (e.g., keys on a keyboard vs the entire keyboard itself). This work uses multi-level masks and makes great progress on the problem of semantic 3D grouping.
This work shows how you can use diffusion to clone faces in images. There are many ways to do this with deep fakes, but diffusion is interesting due to its ability to inpaint other pieces of the image as needed.
Sponsored
Retrieval augmented generation (RAG) might be catching all the AI buzz, but how do you know that your RAG system is actually working?
DiverseEvol is a new method that lets an AI model choose its own training data to get better without needing help from people or other advanced AI systems.
This project explores Self-Rewarding Language Models, where the model acts as its own judge to improve training. By using this method, researchers were able to enhance a model's performance, surpassing other systems like GPT-4.
Depth Anything is a new approach to monocular depth estimation that relies on a large-scale dataset of about 62 million images to improve its accuracy. By using data augmentation and auxiliary supervision from pre-trained encoders, this model achieves impressive generalization capabilities and sets new standards in depth estimation.
This author revised their AI timeline predictions based on advancements in scaling up models. They are now estimating a 10% chance of achieving Artificial General Intelligence by 2028 and a 50% chance by 2045. These changes are attributed to the effectiveness of large language models and the realization that many intelligent capabilities may emerge at scale.
Researchers introduce the "Prompt Highlighter," a method that revolutionizes text generation in multi-modal language models by allowing users to highlight parts of prompts.
Sponsored
Acquiring high-quality datasets is critical for AI and ML. Third-party data - and specifically aggregated news articles - can
Microsoft CEO Satya Nadella is content with the current state of Microsoft's partnership with OpenAI, focusing on stability rather than control amid regulatory reviews in the EU and UK. He emphasizes Microsoft's independent AI developments alongside their significant investment in OpenAI.
MM-Interleaved is a new generative model that excels in processing and creating interleaved image-text data.
Turn your codebase into Wikipedia-style articles using AI.
Since converting to an ETF, the Grayscale Bitcoin Trust has had over $2 billion in outflows, with the FTX bankruptcy estate accounting for a significant portion of that. Additional outflows have partly been attributed to investors moving to lower-fee alternatives like BlackRock and Fidelity, which have attracted significant inflows. Many were calling for the ETFs to be a “sell the news” event, and it has turned out that way in the short term. However, the downward pressure could ease up now that the FTX estate has fully exited its GBTC positions.
Hong Kong could see the launch of its first spot crypto ETF by mid-year, according to local crypto exchange OSL. OSL is reportedly in discussions with multiple fund companies, with five firms leading the pack. OSL views this as a way to make it easier for the average person to gain exposure to crypto and to put protections in place to safeguard investors.
Magic Eden is launching Non-Fungible DAO, which aims to facilitate the adoption of NFT commerce protocols. It will own and manage various open-source technologies for NFTs including smart contracts on Solana, Bitcoin, Ethereum, and Polygon. The DAO's development and protocols will be governed by the token holders, who will also earn most of the token supply.
A new project called Rune Coin launched yesterday. Utilizing a peer-to-peer rune allocation system known as RSICs, 21,000 circuits were airdropped to the Ordinals community. RSIC holders can use them to mine runes, sell them on the market, or keep them. RSICs must be activated to begin mining. Rune allocations vary with flat, random, boosted, and halvening distributions. Game rules and updates can be found on Rune Coin's website and Twitter channel.