中文 EN

Bubble's Brain - 2025-12-15

Daily Brief

AI News 2025-12-15

AI Daily Brief

Summary

OpenAI removed the equity vesting cliff for new hires to encourage risk-taking and innovation.
Adobe integrated Photoshop, Acrobat, and Express directly into the ChatGPT interface for easier editing workflows.

Shenzhen Metro piloted the world’s first AI “guide dog” device for visually impaired riders.
Tnkr.ai is positioning itself as a “GitHub for robotics” by unifying hardware, software, data, and models into one workspace.

DeepMind co-founder Shane Legg estimates a 50% chance of reaching “minimal AGI” by 2028.
OpenAI shared that 85% of Sora’s Android app code was generated by GPT-5.1-Codex, built by a team of four in 28 days.

WeChat Input Method for iOS is upgrading voice input with dialect support and longer session handling.
xAI’s Grok drew criticism after producing major errors and irrelevant claims while responding to breaking news queries.

Today’s AI News

  1. OpenAI removes the equity vesting cliff for new hires: To stay competitive in the talent market and encourage employees to take bigger bets, OpenAI has reportedly eliminated the vesting “cliff” for new hires. Previously, employees needed to work at least six months before receiving their first equity grant. The move is intended to reduce anxiety about being laid off before vesting. Reports also note OpenAI’s equity compensation expense this year could reach roughly $6B, nearly half of its expected revenue.

  2. Adobe brings Photoshop, Acrobat, and Express into ChatGPT: Adobe announced it is integrating Photoshop, Acrobat, and Express directly into the ChatGPT interface. Users can now edit images and documents via simple text instructions—such as customizing photos, designing invitations, or updating a PDF resume—without needing advanced professional skills. Users enable the feature by connecting the relevant apps in ChatGPT settings.

  3. Shenzhen Metro pilots an AI “guide dog” device for visually impaired riders: Shenzhen Metro launched a pilot at the Huangmugang transit hub for an AI assistant device nicknamed “Xiaosuan” (小蒜), described as the first such product in global rail transit. It supports voice commands for route planning, obstacle avoidance, facility recognition, and auto return-to-base. The initial rollout covers the non-paid area outside fare gates with staff support, with plans to expand to more stations.

  4. Tnkr.ai: a “GitHub for robotics” aiming to streamline robot development: Robotics platform Tnkr.ai aims to solve fragmentation across hardware, software, data, and AI models in robot R&D. It consolidates these into a unified workspace, supporting open collaboration, real-time AI documentation generation, cross-platform tool integration, and interactive 3D visualization. The goal is to lower the barrier to entry and create a closed-loop workflow from design and coding to data feedback.

  5. DeepMind co-founder predicts 50% chance of “minimal AGI” by 2028: DeepMind co-founder Shane Legg estimated a 50% probability of reaching “minimal AGI” by 2028—defined as an AI system that can perform most common human cognitive tasks. He suggested that “full AGI” (covering the full range of human cognition) could follow within three to six years after that, and proposed a broader testing framework to assess AI progress.

  6. OpenAI: Sora Android app shipped in 28 days with 85% code from GPT-5.1-Codex: OpenAI shared details of how it built the Sora Android app: reportedly 85% of the code was generated by GPT-5.1-Codex, and the app went from prototype to global launch in 28 days with a core team of four engineers. The team said they manually defined architecture and wrote specification documents to steer the model, highlighting the role of AI in cross-platform code translation and accelerating development.

  7. WeChat Input Method iOS 3.0 beta upgrades voice input with dialect support: WeChat Input Method for iOS started testing version 3.0.0, focusing on voice input improvements. It adds native support for Chinese, English, and multiple Chinese dialects, with automatic detection and transcription. The update also removes voice input time limits, adds “voice standby” settings for faster access, and signals a shift from basic input tooling toward a large-model-powered voice interaction experience.

  8. Grok faces renewed trust concerns after errors in breaking news responses: xAI’s Grok drew criticism for major factual errors when answering questions about a breaking incident at Australia’s Bondi Beach. Reports say it repeatedly misidentified individuals, produced irrelevant geopolitical accusations when users uploaded images, and mixed details across different events—sometimes inserting the incident into unrelated queries. The episode highlights persistent challenges in handling real-time news and fact-checking, and raises further trust concerns.

Comments