AiPulse — Issue No. 52

Mistral aims to build 1 GW of computing capacity by 2030, having raised 3 billion euros led by Samsung Electronics to meet demand for independent AI infrastructure in Europe and reduce dependence on the US.

OpenAI's announcement that it had found an AI-assisted solution to the 200-year-old Navier-Stokes equation became controversial due to claims by Tristan Buckmaster and Levent Alpöge that the company was tracking their own work and attempting to influence credit attribution.

Developed in collaboration between the University of Hong Kong and Hewlett Packard Labs, this chip offers approximately twice the storage capacity of previous designs thanks to an algorithm that compensates for hardware defects, while reducing the number of devices by up to 95%, providing a critical breakthrough for low-power edge artificial intelligence applications.

The new artificial intelligence tool aims to accelerate the process of unraveling the genetic origins of diseases and developing new treatments by predicting the biological effects of billions of possible mutations in DNA.

ShiftLens enables the production of 3D-printed objects that change color through mechanical movement without requiring sensors or complex electronics and provide users with feedback on correct usage.

The restrictions imposed by Chinese manufacturers, such as security reviews and revenue thresholds for commercial use in models like GLM-5.3 and Kimi K3, which contrast with Western companies' adoption of Apache 2.0, are increasing access barriers in the open-source ecosystem.

According to Bloomberg, DeepSeek is preparing to build a one-gigawatt artificial intelligence data center in China's Inner Mongolia region using Huawei Ascend chips due to US export restrictions.

Meta has made its personal AI agent, Muse, which performs tasks such as sending emails and shopping, available for free in the United States.

Adobe announced artificial intelligence tools that generate video, audio, and music without leaving the timeline in Premiere and After Effects, unifying the professional post-production workflow in a single interface ahead of IBC 2026.

Llama.cpp vs vLLM: Which Local LLM Engine Actually Scales?

It compares the fundamental differences between Llama.cpp and vLLM and which is more suitable in terms of scalability for running local LLMs.