Claude Code v2.1.235 Released, Copilot & Google Cloud AI Updates

This official release marks the availability of Claude Code v2.1.235, bringing new features and critical bug fixes to the development tool. A key enhancement in this version is the addition of an optional spellcheck setting. This feature actively underlines misspelled words directly within the prompt input field as a user types, leveraging existing system spellcheckers such as `aspell`, `hunspell`, or `ispell` installed on the user's machine. The update also addresses a significant technical issue: a bug that caused whole-prompt-cache invalidation when a language model was used. This fix ensures more stable and efficient performance, preventing unnecessary cache resets. Users are encouraged to upgrade to this version to benefit from improved usability and reliability when interacting with Claude models through the development environment.

HeyGen has successfully ported its large-scale Avatar IV video generation model, comprising over 18 billion parameters, to Google Cloud's Trillium (v6e) TPUs. This migration leveraged `torchax` and XLA, Google's open-source linear algebra compiler, to optimize performance on the specialized hardware. The implementation utilized advanced parallelism techniques, including Fully Sharded Data Parallelism (FSDP) and Ulysses sequence parallelism, distributed across an eight-chip TPU mesh. This technical achievement resulted in a significant performance gain, with HeyGen reporting a 1.86x speedup compared to their previous setup. The collaboration demonstrates how large, complex AI models can be efficiently deployed and scaled on cloud infrastructure using Google's AI hardware and software ecosystem. This detailed account offers insights into practical strategies for optimizing deep learning workloads for high-performance computing environments.

GitHub Copilot for JetBrains now supports enterprise-managed settings, providing organizations with enhanced control over the AI assistant's deployment and usage. This update allows administrators to implement consistent policies across their development teams, ensuring compliance and standardized configurations. Key areas covered by these new settings include plugin governance, control over access to the MCP server, OpenTelemetry integration, and various permission modes. This functionality enables administrators to centrally define and enforce settings for all users within their enterprise. For instance, they can manage which plugins are allowed, configure how Copilot interacts with backend services via MCP server access, and set up telemetry for monitoring usage and performance. The introduction of permission modes further refines control, allowing granular management of Copilot's capabilities based on user roles or team requirements.