AI News 2025-12-23
AI Daily Brief
Summary
Perplexity and OpenAI both launched annual review features to help users summarize personal activities or AI interactions.
Swedish AI programming company Lovable closed a $330 million Series B at a $6.6 billion valuation; its platform lets non-technical users develop apps through prompts.
GPT-5 independently solved a long-standing math problem, sparking discussion about AI contributions and academic transparency.
Zhipu AI released new model GLM-4.7 with significant improvements in thinking systems and multitasking.
Today’s AI News
Perplexity launches “2025 Year in Review” widget for easy personal activity summaries: Meanwhile, OpenAI also launched a “Your ChatGPT Year in Review” feature, helping users review their interactions with ChatGPT throughout 2025, highlighting conversation topics and usage statistics.
Swedish AI programming startup Lovable announced completion of a $330 million Series B round, valuation soaring to $6.6 billion, tripling within a few months. The round was led by Alphabet’s CapitalG and Menlo Ventures, with participation from NVIDIA and Salesforce Ventures. Lovable’s core technology enables non-technical users to develop apps and websites through text prompts. The platform has attracted知名 clients including Uber and plans to use new funding to strengthen product integration and upgrade infrastructure.
Swiss mathematician Johannes Schmidt recently published results stating that GPT-5 independently solved a long-standing unsolved math problem without human intervention, with its solution demonstrating cross-disciplinary creativity. The proof process is undergoing peer review. Additionally, Schmidt’s paper itself is an experiment on research transparency, where every step of human-AI collaboration is precisely annotated and traced. This approach ensures research integrity but has been questioned by some scholars for potentially hindering innovation due to cumbersome operations, sparking deeper discussion about how to define AI’s independent contributions and future academic publishing standards.
As 2025 draws to a close, OpenAI launched a personalized feature called “Year in Review” for ChatGPT: The feature generates a unique AI interaction profile by aggregating user data. The report includes statistics on total messages sent throughout the year, analyzes chat habits, and generates a pixel-art custom image based on high-frequency conversation topics. Additionally, the system assigns users fun “personality badges” and creative titles like “Producer” or “Navigator” based on their interaction patterns. The feature is currently available in regions including the U.S., U.K., Canada, Singapore, and Australia; eligible users can access it from the app homepage or by sending a command after granting authorization.
Zhipu AI released new generation multimodal and agent model GLM-4.7: The core innovation of this version is introducing a new “thinking system” with three thinking modes: Interleaved Thinking (step-by-step thinking to improve output stability), Preserved Thinking (memory chain-of-thought to reduce logical drift), and Turn-level Thinking (user-controllable thinking depth). This makes the model perform significantly better in complex programming, logical reasoning, and multi-turn tasks. Additionally, GLM-4.7 has substantially improved capabilities in coding, frontend visual design (such as generating modern web pages and presentations), and tool calling (like web search). Its overall performance falls between GPT-5 and Claude 4.5, with advantages in code generation, thinking mechanisms, and visual output, along with high cost-effectiveness.