Bubble's Brain - 2025-12-24

AI News 2025-12-24

AI Daily Brief

Summary

ChatGPT launched "Year in Review" feature, generating personalized AI interaction profiles with usage data and badges.
Seedance 1.5 Pro video model launched on Volcano Ark, supporting audio-visual sync and multilingual dubbing to enhance creation efficiency.
Anthropic open-sourced Agent Skills knowledge base to extend Claude model's capabilities while reducing resource consumption.
Baidu's ERNIE 5.0 model topped international evaluation platform LMArena's text leaderboard, demonstrating outstanding creative writing and complex instruction processing capabilities.
DingTalk launched enterprise AI hardware DingTalk Real, supporting intranet deployment and various office Agents.
MiniMax released open-source coding model M2.1, performing excellently across multiple benchmarks and surpassing closed-source models.

Today’s AI News

  1. ChatGPT launches “Year in Review” feature, generating personalized AI interaction profiles: OpenAI officially launched “Your Year with ChatGPT” feature, generating a personalized year-in-review including full-year usage data, AI personality badges, and custom poems and artwork for eligible users. The feature requires users to proactively enable relevant permissions and meet conversation volume thresholds, currently available only to some English-region individual users. Reports indicate that as adult content features may be opened in the future, year-in-review content may also present a “bolder” appearance.

  2. Seedance 1.5 Pro video generation model launched on Volcano Ark: Doubao’s video generation model Seedance 1.5 Pro officially launched on Volcano Ark, with outstanding performance in audio-visual synchronization, multilingual dialogue dubbing, and cinematic narrative tension. Enterprises can access services via API, while individual users can experience across multiple platforms. The model supports millisecond-level audio-visual sync and native sound effect generation, with plans to launch Draft sample functionality to improve creation efficiency.

  3. Anthropic open-sourced Agent Skills knowledge base, extending Claude model’s capabilities: Anthropic open-sourced “Agent Skills” knowledge base on GitHub, providing 16 exemplary skills covering areas like document processing and creative design. Developers can use them directly or create custom skills based on templates to extend Claude model’s application capabilities. The feature uses on-demand loading mechanism to reduce resource consumption and is available to all paid users.

  4. Baidu’s ERNIE 5.0 model topped the text leaderboard of international evaluation platform LMArena with 1,451 points, surpassing mainstream models like Claude-Opus-4-1 and GPT-5.2, and performing excellently in creative writing and complex instructions. The model has 2.4 trillion parameters and multimodal input capabilities. Its success marks that Chinese AI technology is advancing to international forefront.

  5. DingTalk launched enterprise AI hardware called DingTalk Real, designed to support “Agent” roles within enterprises. The device features intranet deployment capability, can efficiently read internal and public enterprise data, and includes various office Agents to improve efficiency. Product also includes an emergency power-off function to handle uncontrollable situations.

  6. MiniMax officially released open-source coding and agent model MiniMax M2.1: The model performs excellently across multiple coding benchmarks, particularly achieving a high score of 88.6% on VIBE-bench, surpassing multiple leading closed-source models. It reaches advanced levels across multiple programming languages and is considered a powerful open-source model for the agent era.