Custom agents have been introduced in GitHub Copilot CLI to convert single-use terminal prompts into repeatable workflows.

These agents enable the CLI to comprehend specific stack details and team workflows, making processes reviewable and consistent.

Notion uses Codex to enhance its capabilities with AI Voice Input for the web and streamline specification processes.

Codex enables Notion to generate specifications in a single step and significantly boosts the engineering efficiency of small teams. This integration leverages AI to simplify complex tasks and improve productivity.

Anthropic launched Claude Fable 5, a Mythos-class model made safe for general use.

Fable 5 surpasses previous models in AI benchmarks, excelling in software engineering, vision, and scientific research. Safeguards are in place, redirecting some queries to Claude Opus 4.

OpenAI has submitted a confidential draft S-1 to the U.S. Securities and Exchange Commission (SEC).

The timing for any subsequent steps following this submission has not been decided yet.

MiMo and TileRT have released the Xiaomi MiMo-V2.5-Pro-UltraSpeed, achieving a 1000 tokens per second generation speed on a 1-trillion-parameter model using commodity GPUs.

This release marks the first time such a speed has been reached on a model of this size, accomplished through extreme model-system codesign. The UltraSpeed API is available at a promotional price, offering approximately 10 times the generation speed of MiMo-V2.

A software engineer with a decade of experience in web frontend and backend development is facing career challenges due to the rise of large language models (LLMs).

Having specialized in finance-related software development, the engineer recently joined a finance-focused company that encourages the use of AI tools like ChatGPT and Claude Enterprise for coding and research. Despite this, the engineer is required to review and take responsibility for all code used in production.

Google DeepMind released Gemma 4 quantization-aware training checkpoints, which optimize compression for better mobile and laptop efficiency.

These checkpoints use Quantization-Aware Training (QAT) to simulate quantization during training, minimizing quality loss when compressing the model. The release includes QAT checkpoints for the Q4_0 format and a new format for mobile, reducing the memory footprint of Gemma 4 E2B to 1GB. QAT integrates quantization into training, enhancing quality beyond standard Post-Training Quantization (PTQ) while maintaining Gemma 4's capabilities.

Meta disclosed that thousands of Instagram accounts were hacked by exploiting a vulnerability in its AI chatbot.

The flaw allowed hackers to reset passwords on accounts lacking two-factor authentication by redirecting verification codes to their own email addresses. Meta has notified at least 20,225 affected users, including 30 in Maine, about the breach.