Matthias Bastian

Matthias is the co-founder and publisher of THE DECODER, exploring how AI is fundamentally changing the relationship between humans and computers.

Matthias Bastian

Apr 8, 2025

AI Index 2025 documents continued growth in AI capabilities and adoption

The Stanford Institute for Human-Centered AI's annual report is considered the most comprehensive data-based analysis of the AI landscape. It shows how technology is increasingly permeating all areas of life.

Matthias Bastian

Apr 6, 2025

Google adds web search capabilities to NotebookLM

Matthias Bastian

Apr 6, 2025

Microsoft releases real-time AI-generated playable demo of Quake II

Matthias Bastian

Apr 6, 2025

Meta releases first multimodal Llama-4 models, leaves EU out in the cold

Matthias Bastian

Apr 5, 2025

Google expands access to Gemini 2.5 Pro amid strong benchmark results

Google has expanded access to its Gemini 2.5 Pro AI model through a Free Tier option and tiered API pricing, offering lower costs compared to similar models in the market. The model achieved 84% accuracy on the GPQA Diamond scientific benchmark test, performing better than human experts, with results verified by EpochAI. The AI community, particularly developers on X, commends Gemini 2.5 Pro's effectiveness and cost efficiency, especially for programming and coding applications.

Matthias Bastian

Apr 5, 2025

Meta reportedly prepares to launch Llama 4 after delays and internal overhaul

Meta wants to provide a strong counterweight to OpenAI & Co. with Llama 4. However, the model came later than planned - and had to be fundamentally revised both technically and organizationally.

Matthias Bastian

Apr 4, 2025

Microsoft adds ChatGPT-style features to expand Copilot's capabilities

Microsoft is expanding its AI assistant Copilot with new features like "Memory" for creating user profiles, "Actions" for automating tasks on websites, and "Deep Research" for conducting complex research. Copilot Vision, now available on Windows, iOS, and Android, can analyze images and videos on the screen, enabling interactions such as organizing files or adjusting system settings. Additional updates include personalized podcasts, a digital content canvas called "Pages," and "Copilot Search," which combines traditional web search with AI-generated answers, with some features currently in testing.

Matthias Bastian

Apr 4, 2025

OpenAI plans GPT-5 release in "a few months," shifts strategy on reasoning models

OpenAI is delaying the launch of GPT-5 by "a few months" to improve the model's capabilities and prepare the infrastructure for high usage. The reasoning models o3 and o4-mini will now be launched separately, with CEO Sam Altman noting that o3 has significantly improved since its internal test phase. OpenAI has shifted away from its original plan to combine GPT and reasoning models into a single platform, choosing instead to release them as distinct, specialized systems.

Matthias Bastian

Apr 4, 2025

Midjourney launches v7 image model with draft mode for faster iteration

Midjourney has released an alpha version of its V7 image generation model, offering enhanced image quality and introducing personalization as a standard feature for the first time. The new "Draft Mode" generates images ten times faster than the previous V6 model and at half the cost, with lower but consistent quality, making it ideal for quickly developing ideas and creating image variations. According to Midjourney, V7 will initially be available only in Turbo and Relax modes, with additional features such as improved image processing and a new reference system for characters and objects to be rolled out gradually over the next two months.

Matthias Bastian

Apr 3, 2025

Microsoft reportedly pulls back on global data center expansion

The technology group Microsoft has stopped or delayed several data center projects worldwide. This decision raises questions about the actual demand for AI services.