xAI and Cursor Set to Release Joint AI Coding Model as Early as This Week

·2 views

xAI and Cursor are preparing to launch a jointly developed AI coding model, with an internal memo obtained by The Information citing a release window of 'as early as Wednesday.' The unnamed model was trained from scratch on xAI's Colossus supercomputer, with Cursor CEO Michael Truell publicly confirming the development at a recent customer event. Internal testing has been conducted with Tesla and SpaceX employees, and the model is positioned to compete with Anthropic and OpenAI in certain areas, though no public benchmarks have been released. The launch was delayed by a few days to allow engineers to improve the model's efficiency, with a key emphasis on fast information processing suited to real-time coding assistance. The collaboration gives Cursor, which previously lacked the compute resources to train a frontier-tier model independently, access to one of the largest AI training clusters assembled for public-facing work.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

How to Shrink PNG Files Without Visible Quality Loss Using WebP or JPG

PNG is a lossless image format that preserves every pixel exactly, which often results in unnecessarily large file sizes for everyday use. The most effective way to reduce a PNG's size is to re-encode it into a format that supports quality trade-offs, such as WebP or JPG. WebP is the preferred option because it retains transparency support while typically achieving 60–90% smaller file sizes at around 90% quality with no noticeable visual difference. JPG serves as a fallback for platforms that do not accept WebP, though it cannot preserve transparent backgrounds. For best results, always compress from the original PNG rather than a previously compressed copy.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Finds Undocumented Google Patents Endpoint That Needs No API Key

A developer seeking to track competitor patent filings discovered an undocumented XHR endpoint on Google Patents after official sources like the USPTO portal and PatentsView either required API keys or were inaccessible. The endpoint, accessible via a simple HTTP request to patents.google.com, returns structured JSON data including patent titles, assignees, inventors, and publication dates without any authentication. The developer identified several quirks, including a roughly two-month indexing lag, unreliable server-side date filters, and unsorted results within the newest-first view, all of which require client-side workarounds. To handle these issues reliably, a 90-day lookback window and client-side date filtering are recommended over the endpoint's built-in query parameters. The developer packaged the solution as a scheduled automation tool, enabling users to monitor patent activity by keyword or company for competitive research purposes.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How Django's select_for_update Prevents Overselling in E-Commerce Checkouts

Race conditions during checkout can cause multiple orders to be placed for the same last unit of stock, a common problem on moderately trafficked e-commerce platforms. A Django developer addressed this in a single-vendor e-commerce project by separating actual stock from reserved_stock, a field that temporarily holds inventory during an in-progress checkout. Rather than decrementing real stock at checkout, the system only reduces it once payment is confirmed via a Midtrans webhook, preventing phantom availability without risking miscounts from abandoned sessions or failed payments. The solution uses Django's select_for_update() to lock product rows at the database level during validation, forcing concurrent requests to wait rather than read stale data. Consistent row ordering via order_by('id') avoids potential deadlocks, while a single bulk_update call keeps the reservation write efficient.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Calm Technology: The 1995 Design Philosophy That Challenges Today's Notification Culture

Calm Technology is a design concept coined in 1995 by Mark Weiser and John Seely Brown at Xerox PARC, arguing that devices should demand attention only when truly necessary. The philosophy distinguishes between two zones of human attention — active focus and peripheral awareness — and holds that well-designed technology should operate quietly in the background until urgency requires otherwise. Designer Amber Case revived and formalized the principles in 2015, emphasizing minimal attention use and prioritizing calm over anxiety. The concept remains largely at odds with mainstream tech business models, where user engagement and frequent notifications are central to generating revenue. As a result, despite its logical appeal, calm technology has seen little adoption among major device and app makers.

0 comments Read more at DEV Community