Small Drones Spark Giant Cost Crisis In Modern Warfare
The Cost-Exchange Challenge in Modern Warfare: How Cheap Drones Are Reshaping the Battlefield
Modern …
15. September 2025
Cursor Cracks Code With Ai Revolution Boosts Developer Productivity By 48
Cursor, an AI-powered coding platform, has made significant strides in enhancing its Tab model, the autocomplete system that provides suggestions for developers. The company’s commitment to leveraging real-time reinforcement learning (RL) to improve the accuracy and quality of suggestions is evident in this upgrade.
The new iteration reduces low-quality suggestions by 21% while boosting acceptance rates by 28%. This success can be attributed to Cursor’s innovative approach, which involves training a separate model to predict whether a suggestion will be accepted or not. Inspired by a 2022 research study on GitHub Copilot, which demonstrated the effectiveness of using logistic regression filters to predict user behavior, Cursor’s solution employs policy gradient methods.
Policy gradient methods enable the Tab model to learn from feedback and adapt to user behavior in real-time. This approach requires ‘on-policy’ data, collected directly from the model being used. To achieve this, Cursor deploys new checkpoints to users multiple times a day, allowing for quick retraining on fresh interactions. The platform assigns rewards or penalties based on user feedback: accepted suggestions receive rewards, rejected ones incur penalties, and silent choices result in no reward.
This mechanism encourages the model to learn from its mistakes and improve over time. To overcome the challenge of collecting ‘on-policy’ data, Cursor has implemented a system that allows for rapid retraining on new interactions. This process typically takes between 1.5 to 2 hours, although they aim to reduce this timeframe further in the future.
The Tab model is an integral part of Cursor’s platform, handling over 400 million requests per day. Its performance has a direct impact on user experience, and Cursor’s efforts to improve its accuracy and relevance are essential for providing a high-quality coding environment.
Cursor’s use of real-time RL marks a significant milestone in the field, demonstrating the potential benefits of this approach at scale. The platform’s ability to adapt and learn from feedback without human intervention is a testament to the power of AI in improving coding experiences.
A major fundraising round announced by Cursor’s parent company Anysphere in June has secured $900 million at a valuation of $9.9 billion, led by Thrive Capital, Accel, Andreessen Horowitz (a16z), and DST. This investment will undoubtedly fuel Cursor’s growth and expansion plans, including the development of new features and services.
The success of Cursor’s Tab model upgrade highlights the potential for AI-powered coding platforms to revolutionize the way developers work. The introduction of an ‘Ultra’ plan, priced at $20 per month, which promises 20x more usage than the Pro tier, and a platform update with automatic code review capabilities, memory management, and Model Context Protocol (MCP) server setup in a single click, demonstrate the platform’s commitment to innovation.
As the industry continues to evolve, it is likely that we will see more innovative applications of RL and other machine learning techniques to enhance coding experiences. Cursor’s use of real-time reinforcement learning represents a significant achievement in the field, demonstrating the potential for AI to drive improvements in coding accuracy and user experience.