News reporting focused on AI and machine learning, covering the companies behind these technologies, their real-world applications, and the ethical concerns they raise. This includes areas like generative AI (large language models, text-to-image and video), speech tech, and predictive analytics.
OpenAI has merged a pull request on GitHub that reduces the Codex model context size from 372,000 tokens to 272,000 tokens. The change, part of a hotfix for version 0.144, was merged on July 18, 2026, and affects bundled model metadata.
Google announced an update to Google Vids that lets users create custom digital avatars from a selfie and voice recording. The company also brought its multi-modal AI model Gemini Omni to the platform, enabling video creation from text prompts and reference images, along with step-by-step editing. The updates position Vids as a broader video creation tool within Google Workspace, competing with startups like HeyGen and Synthesia.
The European Union has ordered Google to give rival AI assistants and search engines greater access to Android and Google Search data under the Digital Markets Act. The two decisions, announced Thursday, require Google to provide comparable system features and data access to competitors, potentially reshaping how users interact with Android devices and search services in Europe.
Simon Willison accidentally activated a pet in Codex Desktop and then created his own custom pet called Pedalican, a pelican on a bicycle. The process used GPT-5.6 Sol xhigh and gpt-image-2 to generate sprite assets. Willison documented the entire process, including generated images, sprite sheets, and GIFs, and shared the open-source skills on GitHub.
Noma Labs discovered a critical prompt injection vulnerability in GitHub's new Agentic Workflows, dubbed GitLost. An unauthenticated attacker could leak private repository data by posting a crafted GitHub Issue in a public repository belonging to the same organization. The vulnerability exploits the AI agent's failure to distinguish trusted instructions from untrusted user content.
Microsoft's shift to per-token pricing for GitHub Copilot has been dubbed the 'Tokenpocalypse' by Reddit users. On TechCrunch's Equity podcast, hosts discuss how this change signals broader cost pass-through in the AI industry, the rapid rise and fall of tokenmaxxxing, and the challenges AI companies face as they prepare for IPOs amid escalating costs and regulatory uncertainty.
A developer has released tiny-vllm, a high-performance LLM inference engine written in C++ and CUDA, alongside a comprehensive course that walks through implementing the engine from scratch. The project covers loading models from Safetensors, running prefill and decode, building CUDA kernels, KV cache, and more. It is intended as a learning resource for engineers and educators.
GitHub is testing an experimental accessibility agent integrated with GitHub Copilot CLI and VS Code to answer accessibility questions and fix simple issues in front-end code changes. The agent has examined 3,535 pull requests with a 68% fix rate, targeting common problems like clear structure for assistive tech and text alternatives. Teams share insights on agent design, sub-agents, limitations, and using past issue data to improve results.
Microsoft plans to revoke most internal licenses for Anthropic's Claude Code and shift developers to its GitHub Copilot CLI tool. The move affects the Experiences and Devices team handling Windows and other products, with usage ending by late June. Sources point to cost savings at fiscal year-end alongside strategic reasons.
A detailed lab test shows how CVE-2026-31431, known as Copy Fail, escalates privileges inside rootless Podman containers but fails to escape to the host due to user namespace mappings. The analysis covers shellcode disassembly, exploit tracing, and kernel behavior via eBPF. Rootless setups on GNOME GitLab runners contain the issue effectively.
A GitHub project provides a hands-on workshop for building a full GPT training pipeline. Users write code for tokenization, transformer architecture, training loop, and text generation to train a 10 million parameter model on a laptop in under an hour. Designed for a single session, it uses character-level tokenization on Shakespeare data and supports Apple Silicon, NVIDIA GPUs, or CPU.
Microsoft included a 'Co-Authored-by Copilot' line in Visual Studio Code Git commits, even when AI features were turned off. A product manager pushed the change, which faced backlash on GitHub and Hacker News. The developer plans to revert it in version 1.119.