Recent AI Developments in September 2025 | Generated by AI
Here are some of the most recent developments in AI as of mid-September 2025:
🔍 Key AI News Highlights
-
Meta licensing talks with publishers Meta is in discussions with major media companies (Axel Springer, Fox Corp, News Corp) to license news content for its AI tools. This is part of a larger trend: AI companies securing formal agreements with publishers, reflecting concerns over content usage, copyright, and monetization. (Reuters)
-
DeepMind’s “Kasparov moment” in coding Google DeepMind’s AI model, Gemini 2.5, solved a complex programming problem at the ICPC World Finals that human teams couldn’t solve, earning a “gold medal” level recognition. It shows increasing strength in abstract reasoning and problem solving. (The Times)
-
AI predictions for health outcomes A new tool called Delphi-2M can predict someone’s risk of over 1,000 diseases, over up to 20 years, using data like medical history, lifestyle, demographics. It’s been validated using UK and Danish biobank data. Big implications for personalized preventive medicine. (The Guardian)
-
Optical generative model to reduce energy use Researchers at UCLA developed a hybrid model using optical computing (light) instead of purely electronic computation to generate AI images. The optical component greatly reduces energy usage, which could help with sustainability and edge or wearable devices. (Popular Mechanics)
-
US-China AI race & regulation dominate discussions At the Axios AI+ DC Summit, speakers emphasized AI regulation, geopolitical competition (especially chip dependency), and the impact of AI on jobs. There’s increasing agreement that oversight, safety standards, and trust need more attention. (Axios)
-
AI security is being prioritized CrowdStrike is acquiring Pangea Cyber (~US$260 million), a company specializing in securing AI systems, especially against prompt injection attacks and other vulnerabilities. This reflects growing awareness that as AI usage expands, the security risks rise sharply. (The Wall Street Journal)
-
Deterministic behavior in LLMs Mira Murati’s Thinking Machines Lab is working on making large language models deterministic: i.e. ensuring “same input → same output.” Non-determinism (even under identical inputs) can undermine reliability and trust, so this effort aims to make AI responses more predictable. (The Times of India)
-
Governments adopting AI, but scale-up is hard An OECD report notes governments are increasingly using AI in public administrations, but many projects are stuck in pilot phases. Barriers include trust, skills, regulation, and deployment of AI at scale. (OECD)
-
Intel & NVIDIA collaborate on next-gen AI infrastructure They’re partnering to build new data center and client hardware using custom CPUs/GPUs and leveraging NVIDIA’s NVLink. Big investment in infrastructure for both enterprise and consumer-level AI performance. (Newsroom)
If you like, I can pull up recent AI news specific to Taiwan (or Asia) so it’s more locally relevant. Do you prefer that?