September 2022 Summaries

The new feature in Replit allows users to customize the layout of their Workspace by dragging and dropping tabs or panes into different quadrants. This enables a high degree of flexibility and customization, making it easier for users to organize their screen according to their working style. The design principles behind this update prioritize predictability, explicit state visualization, immediate visual feedback, and direct manipulation, aiming to create an intuitive and learnable interface. The system is built on top of existing patterns, such as traditional tile-based windowing, but with a focus on ease of use and fluidity. Future improvements are planned, including natural language interaction, layout automation, keyboard shortcuts, and more. Overall, this update aims to make Replit feel more powerful and useful for users, while also providing a flexible and customizable workspace that adapts to their needs.

In this blog post, the authors discuss how they built a custom load balancer to improve the performance and scalability of their infrastructure. The need for an alternative load balancer arose due to issues with the Google Cloud Load Balancer (GCLB), which was causing uneven distribution of requests across machines, leading to overloaded and underutilized resources. The authors built a proof-of-concept using subsetting, a simple concept that places requests in the least-loaded machine at any point in time. They compared their results with GCLB and were surprised by the differences. After building a simulator to test different strategies and tweaks, they deployed their custom load balancer in production and found significant improvements in resource utilization uniformity and reduction of spurious errors. The authors are now working on adding vertical scalability and are excited about the potential of their new primitive.

We've learned to build and host large language models by tweaking parameters such as temperature and frequency penalty to control the model's output, using techniques like cycle detection, sampling, layering another model, training to break loops, and finetuning on user feedback to filter out nonsense. We also use benchmark datasets and A/B testing to evaluate performance and make improvements. To speed up training and inference, we've adopted faster transformer architectures, knowledge distillation, and quantization techniques. Effective prompt engineering is crucial, as prompts change with context and can be generated by other models. Deploying the model on GPUs in the cloud requires careful consideration of resource utilization and monitoring to ensure optimal performance and user experience. By tracking system-level metrics and storing user feedback data, we fine-tune our models for better quality and consistency.

We are excited to announce the launch of Replit's 100 Days of Code, a 15-minute daily coding course that will help you build real-world projects and develop a daily coding habit. You will learn from master teacher David Morgan and be part of a vibrant community with collaborative features, live support, and a unique opportunity to publish your code and engage with others. This course is designed for all coding levels, from beginners to advanced coders, and offers a fun and engaging way to improve your coding skills and potentially build profit-generating apps and websites.

Replit has introduced a new history user interface aimed at improving the way programmers manage code versioning and history. Unlike traditional version control systems like git, which require intentional commits and can be complex for some users, Replit's system logs operational transformations (OTs) automatically during every edit, providing a seamless and detailed account of a file's historical states. This feature allows users to non-destructively browse, compare, and restore any specific version of their code, enhancing their workflow without additional mental effort. The history UI includes a playback feature, enabling users to watch their coding process like a movie, and supports dark mode, mobile devices, and a polished interface for ease of use. Replit intends to further develop social and sharing functionalities for this innovative tool.

Replit's Trust & Safety team works behind the scenes to maintain a safe and welcoming community for users, with a focus on supporting freedom of expression and creativity. Moderation is handled through a combination of automated tools and human moderators, who only step in when a possible violation of Replit's Terms of Service has been reported or detected. Users can report concerns using the "kebab menu" and various reporting options, which help to detect and remove inappropriate content, including images with NSFW material. Replit also offers extra safety features for school-age users, such as strict privacy standards and a Firewalled product that prevents external Internet content from being accessed. The platform's Terms of Service have been updated to clarify what types of content are not allowed, and provide support resources for users who may be struggling with thoughts of self-harm or other issues. Overall, Replit aims to stay ahead of malicious actors through ongoing investment in safety measures, ensuring a trusted and safe community for its users.

Replit has introduced Ghostwriter, an AI-powered feature integrated into its IDE that aims to transform how coding is done by offering real-time code completion, generation, transformation, and in-editor code search. Building on the success of Replit's Multiplayer Mode, Ghostwriter acts as an AI pair programmer, leveraging advanced machine learning models to provide code suggestions and explanations, making coding more efficient and accessible, particularly on mobile devices. The development of Ghostwriter involves optimizing open-source large language models to reduce latency and enhance user experience, incorporating strategies such as knowledge distillation and heuristic filtering. This innovation is poised to significantly alter the landscape of software development by making programming more intuitive and accessible, aligning with Replit's vision of empowering the next generation of software creators. Ghostwriter is currently in closed beta, with plans to expand its availability and affordability, much like Replit's Multiplayer Mode.

Lily, a community intern at Replit, reflects on her eight-month internship, which marked the culmination of her active involvement in the Replit community over the past two years. Initially hesitant, Lily applied for a design role at Replit in late 2021 after gaining confidence from her community contributions and hackathon performances. Though she began her role as a Community Intern in January 2022, she engaged in a wide array of tasks, from moderation and organizing events to collaborating with influencers and launching a mentorship program. Her efforts also included content creation and facilitating community interactions. A significant highlight was her involvement in the successful Replit India meet-up, which provided her with her first opportunity to travel and meet online friends in person. Lily expresses gratitude to her mentors and team members for their support and highlights Replit's inclusive, flexible, and learning-friendly environment. As her internship concludes, she remains open to new opportunities in community or design roles while continuing her engagement with Replit as a representative.