OpenAI has released a new blueprint for the US government, drawing a historical comparison between AI and the development of the automotive industry. Although the automobile was invented in Europe, OpenAI argues that the US became the center of the global automotive industry through a more innovation-friendly approach. An example of a negative regulation is the British Red Flag Act, which required someone to walk in front of cars with a red flag. The US, on the other hand, connected private sector visions and innovations with public interests and invested in better road infrastructure.
The blueprint calls for uniform, less restrictive regulation at the federal level. OpenAI warns against losing AI supremacy to China and criticizes existing regulations such as the Chips Act. Sam Altman, CEO of OpenAI, identifies chips, data, energy, and talent as key factors for success in the global AI competition. The document focuses exclusively on Large Language Models and calls for easier AI access for government agencies to increase efficiency, mandatory watermarks for AI-generated content, and a ban on AI-generated CSAM content. Additionally, AI companies should have access to classified computer clusters for security assessments.
The US government plans to control the global trade of AI accelerators with its Export Control Framework for Artificial Intelligence Diffusion. The proposal, which could take effect under the Trump administration after a 120-day comment period, envisions a multi-tiered system. Only 18 privileged countries, including Germany, may purchase advanced GPUs without restrictions. Private users and individual gaming graphics cards are exempt from the restrictions. A maximum of 7 percent of a company’s global computing capacity may be located abroad.
For other countries, strict limits apply: With NVEU status (National Verified End User), up to 320,000 GPUs can be purchased in two years. Without status, there is a limit of 50,000 GPUs per country. Government agreements can double these limits. Orders under 1,700 GPUs are exempt from the restrictions. Additionally, the US plans to control the spread of large AI models developed with US technology. This affects models with 100 quadrillion computing operations during training, equivalent to about two months of computing time on modern supercomputers. Open-source models are exempt from these restrictions. These measures particularly aim to restrict China’s access to advanced AI technology and secure US dominance.
Banks are planning to use AI instead of personnel. According to an analysis by Bloomberg Intelligence, banks will replace about three percent of their jobs with AI over the next three to five years. Employees in customer service and business processes are particularly affected. Large banks like JPMorgan and Goldman Sachs expect job cuts of up to ten percent. At the same time, the institutes expect a profit increase of 17 percent by 2027, equivalent to 180 billion US dollars.
Alibaba has launched a new free chat interface called Qwen Chat. Users can interact with various AI models, including the Qwen2.5-Plus language model and the Qwen2-VL-Max vision-language model. The platform allows uploading documents and images. In the future, web search, image generation, and voice mode will be added. As with all Chinese AI models, government-critical content is censored. Most Qwen models are available as open source under the Apache 2.0 license.
The start-up Astral wants to flood Reddit with AI agents that automatically create marketing posts around the clock. The agent only needs a few pieces of information to autonomously navigate Reddit and act like a human. The AI agent finds a suitable subreddit and formulates a human-sounding advertisement. It seems unnecessary to have a human check the posts. Platform operators like Reddit, X, and Instagram prohibit the use of their APIs for automated AI content. Captchas do not stop the AI bots. Currently, there are no means for platform operators to prevent AI agents. Astral is currently in a closed beta phase. Extensions for Instagram and TikTok are to follow.
The French start-up Mistral has updated its code model, Codestral. Version 25.01 works twice as fast as its predecessor and supports over 80 programming languages. In benchmarks, the model surpasses other code models under 100 billion parameters. In the Copilot Arena, it currently ranks second. Codestral 25.01 is available through various partner platforms and APIs. Users can test the model for free with VS Code or JetBrains.
The start-up Syntilay is bringing the first commercially available AI-designed shoes to market. The slides, in a Crocs style, are custom-made using 3D printing and cost $150. The design process used various AI tools: Midjourney for concept images, Vizcom AI for 3D models, and generative AI for patterns. The fit is determined by a smartphone scan. The project is led by 25-year-old Ben Weiss, supported by Reebok co-founder Joe Foster. Production is initially limited to a few thousand pairs. In the future, other brands could also offer AI-designed shoes through the platform.