I’m welcome Ai decoded, Fast companyWeekly Bulletin violating the most important news in the AI world. You can register to get this newsletter here every week.
Nvidia’s Jensen Huang: The world has been completely wrong
This week in San Jose in Nvidia’s GTC Development event, CEO Jensen Huang called him AI’s “Super Bowl”. Indeed, NVIDIA is a chip supplier of the AI boom and is the most influential (and of course the most profitable) company, which grows. As every year, the company used the main part of the HUANG to announce the next generation chips that will strengthen the training and operation of AI models in the near future.
Huang presented NVIDIA’s first special CPU, Vera, Vera, Vera and two new Rubin “GPA” Vera Rubin “. (The platform is named after astronomer vera Rubin, which reveals the evidence of the dark matter.) Nvidia, Vera is twice as fast as the goal is based on the goal in the Blackwell platform last year. Together, Vera and Rubin double up the fruitless performance of Blackwell.
Huang also revealed Blackwell Ultra, a more powerful version of the current flagship GPU, increased reporting and memory. It is expected to start in the second half of 2025.
These platforms – especially designed to increase performance during vera Rubin-inference, real-time mode justification is played to create a response to AI models. Huang stressed the growing needs of chips that can quickly process and maintain a massowing amount of information during the inference. In a statement on a sign, “Almost the whole world was wrong,” said DeepSeek phenomenon claimed that China could grow the highest levels with less, weaker GPU. “Agent AI, the calculation amount we need as a result of justification, this time as a result of thought, this time is 100 times more than we need this time,” he said. In other words, the demand for the result is to increase and with it, NVIDIA devices.
Huang has also updated the desktop AI platform for project figures, developers, researchers and students. Sent in two versions: GSTX station is a larger GTX station that is capable of working in compact GTX sparks and more complex models. Nvidia says that both of them will reach this summer market.
Like the role of NVIDIA to the Generative AI, it is pushed to the infrastructure and application layer. For this purpose, the company announced a new family of open source, medium-sized justification models called Nemotron. The methane is built on Llama, these models can spark or station office or work in the cloud. Although the most advanced, it is optimized for lighter enterprise tasks, and NVIDIA chips-proposal is used effectively on another promotion to stay within the NVIDIA ecosystem.
Big AI, ‘AI Action Plan’ gives Trump Management Points
When the US government considers the approach to artificial intelligence, the consensus appeared in the second Trump administration is clear: minimal control, maximum freedom for the industry. The theory of the work is that Trump and Vice President JD Vice will receive an approach to the AI regulation, requiring rapid innovation over security mandates.
Both Trump and Vance, US company’s transparency mandates or security rules, claiming that this freedom is important for global leadership, he stressed his work without security rules. “AI does not intend to manually punish the future of the future,” he said. On the first day, the leadership, mainly voluntary Biden-Era AI canceled the rules of security and transparency.
The Trump team has developed its EU policy. As part of these efforts, the Office of Science and Technology Policy (OSTP) asked for public access to the “AI Action Plan”. The plan is to determine the “unnecessary heavy requirements for the” unnecessary heavy requirements “on private sector updates, the plan is aimed at determining” the necessary priority policy measures to increase the AI. Comments clearly accepted what they presented from March 15 and several large AI-Openai, Anthropic, Meta and Google.
In the Board, these companies have noticed that the rules of extreme aggressive security can lead to China’s return in the US race. The most supported exports, such as China, protect against the fight against advanced AI chips to rival nations. Openai, Anthropic and Google, called on the public to protect the “fair use of use” of the Copyright Law, which allows the public to prepare open information to the public.
Again, some comments have warned in terms of more risk. The anthopic government called for creating mechanisms to assess the fact that private sector models can disrupt national security threats and opened special communication channels between intelligence agencies and major AI laboratories.
Interestingly, Openai prefers to ban the Models of Deepseek, China’s starting Deepseek, citing National Security Concerns, which leads to the National Security Concerns, which leads to Huawei equipment. “DeepSeek looks at the requirements of Chinese laws in accordance with the requirements of the user and uses it to develop more skilled systems for the use of CCP,” he said.
Although the AI action plan is written, companies are not waiting for those who wait for serious security rules under existing management and congress. AI systems grow stronger, so take the risks of general scale, potentially unregulated abuse.
AI Training Information Company CEO CEO SAMA SAMA, the lack of workers, may be bad for work for a long time. “It is a good example than means vehicle safety standards,” he says Fast company. “We simply protect the stakeholders when we do not look like ‘extreme regulation’ that stains their seat belts, automotive innovation.”
Superhuman is a work on the right track to integrate AI to apps
Most of the post-ChatGPT AI BOOM has focused on models, but increasingly varies because the EU application is applied and in real world applications. Some applications are completely new, new, which is possible only by a generative AI. Others, like superhuman, think about how technology combines existing products.
Superhuman has been around since 2014 and added AI features Superhuman’s spare and bright design did not change much. His creators integrated AI features purposefully and secretly. Superhuman founder and CEO Rahul Vohra, when we met the Humanns Conference last week, some of the new AI features gave the run.
AI works as a scenic assistant who knows the context and context of your inbox. Provides summaries for long or complex topics. You can prepare answers to your writing, respond to urgent or high-priority messages and adjust the communication samples of contacts to your answers for maximum visibility. It also helps you organize your Inbox by automatic redirecting marketing, social updates or colds to separate folders such as less important email.
Vohra claims can save superhuman users to four hours a week. An internal research by a large consulting firm, 3.3 hours of weekly, the response period accelerated 3.6 hours and increased the email transmission by 60%.
Between $ 30-40 per month, superhuman is not cheap. However, for experts face-to-face, the increase in productivity with the AI may be worth the price.
More AI coverage Fast company:
- The most innovative companies of artificial intelligence for 2025
- AI data centers are working hot. These light engines cool them with less power
- Hollywood warns the AI industry to push the copyright law to change the law
- The rise of the AI manager
Want an exclusive report and trend analysis on technology, business innovation, future and design? Register for Fast company The reward.