LLMs like ChatGPT have limitations, like hallucinations, lack of confidence estimates, and the absence of citations. One approach to address these limitations is to bootstrap consistent LLMs by training them on carefully curated datasets that exclude contradicting information. This could be achieved by starting with a small, high-quality corpus and gradually expanding it using the base model to classify new text documents as consistent or inconsistent.
Xata offers a sustainable free tier for its Postgres and OpenSearch database services by utilizing shared clusters that efficiently host multiple databases. It manages costs effectively through careful cluster management and automation, distinguishing between active and inactive databases to optimize resource allocation.
NVIDIA is transitioning fully towards open-source GPU kernel modules in the upcoming R650 driver release.
USPS shared postal addresses with Meta, LinkedIn, and Snap through hidden tracking pixels, but it stopped after an investigation by TechCrunch, claiming the company was unaware of it.
Coding exercises in technical interviews can be unfair to those without ample practice time outside of their jobs, like those with families and kids.
Microsoft researchers created SpreadsheetLLM, a tool that uses large language models to analyze and manipulate spreadsheet data, and SheetCompressor to address the token constraints of LLMs.
Notion sped up its web browser application by implementing SQLite caching using WebAssembly. It faced challenges with cross-origin isolation and data corruption due to concurrency issues when multiple tabs accessed the database simultaneously. Notion resolved these issues by developing a SharedWorker architecture where only one tab actively interacts with the SQLite database while others can still make queries. It also addressed regressions like slower load times and performance issues on slow devices by loading the WASM SQLite library asynchronously and implementing a “race” mechanism between SQLite and API requests.
The first step to building a well-designed GraphQL schema is to identify the core types in your domain, like Company and JobListing, and then define the relationships between these types. Next, plan query and mutation operations that clients can perform on your API, such as creating or updating job listings. Finally, document your schema thoroughly and use tools to visualize the relationships in your schema.
Virtual threads in Java do not offer a clear advantage over Open Liberty's existing autonomic thread pool for typical cloud-native Java workloads. Open Liberty is an open-source, modular, cloud-native Java application runtime. In CPU-intensive workloads, virtual threads perform worse, with lower throughput than Open Liberty's thread pool. However, virtual threads do ramp up faster from idle to maximum throughput compared to Open Liberty's thread pool due to its thread-per-request model.
Persisted queries are essential for a solid GraphQL API. Authorization is a challenge regardless of API style. While there is a need for `dataloader` to address performance concerns, the claim made by others that authorization problems are unique to GraphQL is untrue. GraphQL encourages continuous evolution, not breaking changes. It allows for precise tracking of API usage.
Python's datetime module can cause confusion when converting between timestamps and datetime objects due to timezone issues. The time.time() function returns a timestamp relative to epoch (UTC), while datetime.datetime.now() returns a timezone-naive object in the local timezone. This discrepancy can lead to incorrect results when converting timestamps back to datetime objects on systems with non-UTC timezones.
This developer, who completes Advent of Code in a new programming language every year, chose Zig for 2023 and found it to be a modernized C with a unique approach to error handling and type manipulation. Zig's standout features include detectable illegal behavior and comptime, which offer improved error detection and type-level programming capabilities respectively. However, Zig could benefit from enhancements in developer experience, including a better language server, clearer error messages, and improved documentation.
Google Docs has introduced new features to convert Markdown to Docs content and vice-versa, export Docs as Markdown, and import Markdown files as Docs.
2024.07.18·workspaceupdates.googleblog.com·1 min read·TLDR Web Dev Magic CLI is a command line utility that uses LLMs to help you use the command line more efficiently.
This is a CodeMirror component for React. CodeMirror is a code editor component for the web, with a rich, comprehensive API that is easy to configure. This React-specific component supports React hooks, TypeScript, and more.
The tech job market is facing a downturn due to rising interest rates, which affect different types of tech companies differently. On top of that, the hiring process in tech is often broken, relying on standardized practices like coding tests and behavioral interviews that may not accurately assess a candidate's abilities or experience. Companies often have unrealistic expectations and job descriptions, leading to burnout and high turnover.
David Buchanan bought a Rabbit R1, a device that has been criticized for its poor performance. He developed a tethered jailbreak that gives root access to the device without making any permanent changes. He also found that the device logs extensive user data, including GPS location, WiFi network name, and nearby cell tower IDs. While Rabbit Inc. has since addressed the logging issue and added a factory reset option, Buchanan is still releasing his jailbreak tool to help other researchers analyze the device.
The JPEG format became the dominant way to share digital photos online due to its lossy compression, which allows for smaller file sizes and faster transmission over the internet. Unlike the GIF, which was created by a single person, the JPEG is a complete standard designed to meet a range of industry needs. The format's success is due to its ability to balance the need for photorealistic images with the requirement for efficient delivery in limited bandwidth environments.
Cloudflare's latest report reveals that nearly 7% of internet traffic is malicious, driven by events like wars and elections, with DDoS attacks being the primary weapon of choice.
The term "IC" (individual contributor) has negative connotations of laziness and lack of ambition.
Starting today, the npm registry will begin removing README content from package version metadata to reduce the size of package packuments and improve the performance of the registry and package managers, including the npm cli.
Sponsored
Does shipping digital experiences quickly mean compromising on quality? Not necessarily. AI and visual development create new ways to build, iterate upon, and optimize applications and websites. This live session will cut through the hype [4] and provide a realistic view of how traditional CMS architecture misses the mark and how emerging technologies address long-standing challenges:
This developer rebuilt a 10-year-old iOS app using Expo, adding Android and TV support along the way. They successfully replicated the original app's functionality using Expo and React Native packages, avoiding the need for custom native code. The new UI design used responsive layouts for different devices, and the app was built and submitted using EAS Build and Submit.
React projects often suffer from messy architecture due to the framework's unopinionated nature. This article focuses on domain logic, which is code that operates on domain models like user objects. The goal is to isolate domain logic from components and move it to a dedicated domain layer, improving code readability, testability, and reusability.
German Strings, designed for data processing, are a custom string type that optimizes for short, immutable strings. Unlike C or C++ strings, German Strings store short strings directly in place and include a prefix for faster comparisons. Long strings are stored with a pointer, length, and prefix, but without extra capacity for appending. German Strings also have three storage classes: temporary, persistent, and transient.
You should review your own pull requests (PRs) in the GitHub UI to catch mistakes and improve code quality before involving other team members. This practice helps identify issues that might be overlooked in the code editor and allows for more productive and focused code reviews. Add comments to PRs to explain design choices and alternative solutions tried before sending them out for review.
Story points, used to estimate effort in Agile development, are unreliable due to their relative nature and inherent variability. Velocity, which converts story points into time, further worsens this problem by creating unrealistic deadlines and discouraging teams from exceeding expectations. A better approach is to break down projects into smaller tasks, creating queues, and tracking the average task rate, which provides a more reliable way to project completion times and adjust for scope changes.
Engineering teams often face a problem when choosing system design tools: they focus on the solution (diagramming tools) rather than the underlying problem (understanding and communicating complex systems). System design is a continuous process that goes beyond diagrams and includes documenting requirements, architecture, and decisions. Modern software systems are too complex for traditional diagramming tools, which lack real-time updates, collaborative features, and the ability to manage cloud resources.
Codestral Mamba is a new open-source model designed to improve code productivity. Unlike Transformer models, it has linear time inference and the ability to model infinite length sequences. It can be used as a local code assistant and deployed using the mistral-inference SDK or TensorRT-LLM.
The AWS Secrets Manager Agent is a client-side HTTP service that can be used to standardize the consumption of secrets from Secrets Manager across environments such as AWS Lambda, Amazon Elastic Container Service, Amazon Elastic Kubernetes Service, and Amazon Elastic Compute Cloud. The Secrets Manager Agent can retrieve and cache secrets in memory so that your applications can consume secrets directly from the cache. That means you can fetch the secrets your application needs from the localhost instead of making calls to the Secrets Manager.
Eidos is an extensible framework for managing personal data throughout its lifetime in one place. It runs in the browser and has offline support and local AI features.
This author initially implemented a bignum library using base-10 digits, but it was slow. They then optimized it by switching to 30-bit digits, which improved memory efficiency and arithmetic operations. This change resulted in an 83% faster addition and a 99% faster multiplication.
AI Lab is a pre-production framework at Meta that continuously A/B tests common ML workflows to proactively improve and prevent regressions on Time to First Batch (TTFB), a key metric for ML engineer productivity.
Andrej Karpathy has announced a new AI+Education company, Eureka Labs, which aims to create an AI-powered learning platform where students can learn from high-quality course materials guided by AI Teaching Assistants.
Small, focused pull requests that do exactly one thing each are easier to review, less likely to have merge conflicts, easier to rollback if something goes wrong, and lead to more accurate descriptions and increased code coverage.
Jon Kern, co-author of the Agile Manifesto, criticizes a recent report on Agile failure rates, arguing that it misrepresents Agile principles and promotes excessive processes that obscure the original spirit of the Manifesto.
An ex-Coinbase employee discusses performance challenges faced when using React Native at Coinbase, including re-rendering issues and solutions implemented, as well as platform-specific optimizations between iOS and Android.
Remix's "Fog of War" feature dynamically loads route information based on user interactions. This optimizes React Router SSR applications by decoupling rendering from fetching, resulting in a "fetch then render" approach that eliminates network waterfalls. It uses a Route Manifest to prefetch route data and components so that navigation is smooth.
Trigger encountered reliability and performance issues in its Node.js application due to event loop lag, which led to high CPU usage and crashes. To fix this, it fixed inefficient code, added pagination, and monitored event loop lag with OpenTelemetry. Moving forward, it will keep an eye on larger payloads, as a learning from this experience.
Booking.com scaled its customer review system by sharding reviews across multiple servers using the internal ID of accommodations as the partition key. When it needed to add more machines to the cluster due to increased traffic, it used the Jump Hash Sharding algorithm to minimize the amount of data that needed to be rearranged. This algorithm guaranteed monotonicity, meaning that data was only transferred from old shards to new shards during resharding, avoiding unnecessary data movement.
It can be hard to maintain focus, especially when switching between tasks. Keeping a work journal to record your thoughts and actions can help you think more clearly and regain context faster. Writing down what you're doing and reading the last few sentences when resuming work can be a helpful tool for improving focus.
Ghost jobs are job postings that companies don't intend to fill. Companies post ghost jobs for various reasons like wanting to hire eventually, maintaining a talent pool, or creating a false impression of company growth. To avoid ghost jobs, job seekers should be wary of vague job descriptions, old listings, and companies with little online presence.
Sponsored
Discover SurveyJS [9], an open-source JavaScript library suite for secure form creation and data collection in your application. Build complex JSON-driven surveys and forms quickly and without manual coding. Collect and store responses while retaining full control over your data. Integrate with any backend system. Enjoy a configurable drag-and-drop interface and a CSS theme editor for precise form
SCALE is a GPGPU programming toolkit that allows CUDA applications to be natively compiled for AMD GPUs. It does not require the CUDA program or its build system to be modified.
Claude Dev is a VSCode extension that can handle complex software development tasks step-by-step. It can read & write files, create entire projects from scratch, and execute terminal commands (after you grant permission). Claude Dev can assist you in ways that go beyond simple code completion or tech support.
PR Pilot is an AI agent for your development workflow that can search and manipulate the code base, browse the internet, and interact with GitHub issues and pull requests.