Smart Architecture on Raw Account: Deepseek breaking up the “bigger is the best” to develop artificial intelligence

The narration of artificial intelligence has reached a critical turning point. Deepseek’s penetration-achieving the latest performance without relying on the most advanced chips-what many announced in NeuPs in December: The future of artificial intelligence is not related to more work with humans and our environment.

As an educated computer scientist in Stanford witnessed the promise and risks of developing artificial intelligence, I see this moment more transformative than the appearance of Chatgpt. We enter what some call the “renaissance of thinking”. Openai’s O1 and Deepseek’s R1 and others move in the brutal hall that is heading towards something more intelligent-doing this unprecedentedly.

This transformation cannot be in time. During his nerve keyword, chief scientist in Openai Elya Sutskever Declare “The gradual will end” because during energy growth, we are bound by limited internet data. Deepseek is validated by this perspective – researchers at the China Company made a similar performance for Openai’s O1 in a small part of the cost, indicating that innovation, not only the power of raw computing, is the way forward.

AI advanced without huge training

World models ascend to fill this gap. World Labs recently He raised 230 million dollars To build artificial intelligence systems that understand reality like humans parallel to the Deepseek approach, where their R1 model is presented “AHA!” Moments-stop reassessing problems just as humans do. These systems, which are inspired by human cognitive processes, are the transformation of everything from environmental modeling into the human interaction.

We are witnessing early victories: Meta’s last update Ray Ban Smart Glasses Continuous contextual conversations with artificial intelligence assistants allow without waking words, as well as translation in the actual time. This is not just a feature of features-it is a preview of how to enhance human capabilities without the need for huge pre-trained models.

However, this development comes with accurate challenges. While Deepseek has greatly reduced the costs through innovative training techniques, this efficiency achievement can lead to an increase in resource consumption in general – a phenomenon known as the name Jevons ParadoxAs technological efficiency improvements often increase resource use rather than low resources.

In the case of artificial intelligence, the cheapest training may mean more models that are trained by more organizations, which may increase net energy consumption. But Deepseek innovation is different: by showing that advanced performance is possible without advanced devices, it does not make artificial intelligence more efficient-it is mainly changed how we deal with the development of the model.

This shift towards smart architecture can help us on the power of raw computing to escape from the Jevons Paradox trap, where the focus moves from “How much can we bear?” To “How can we design our systems?” Professor Ucla Guy Van Den Brock, “The total cost of thinking about the language model is not decreased.” The environmental impact of these systems remains great, pushing the industry towards more efficient solutions – exactly the type of innovation represented by Deepseek.

Give priority to effective structures

This transformation requires new methods. Deepseek is achieved from the fact that the future is not related to building larger models – it is related to building more intelligent and more efficient models that work in harmony with human intelligence and environmental restrictions.

The head of Amnesty International in Mita Yan Lacon Imagine future systems Eliminate days or weeks thinking through complex problems, as humans do. The Deepseek’s-R1 model, with its ability to stop and reconsider, represents a step towards this vision. Despite the intensity of resources, this approach may result in penetrations in climate change solutions, health care innovations and beyond. But as Carnegie Mellon Amit Magazine Wisely warn, we must ask anyone who claims certainty about the place where these techniques will lead us.

For institution leaders, this shift provides a clear path forward. We need to give priority to effective architecture. One can:

• Spreading chains of specialized artificial intelligence agents instead of individual huge models.
• Invest in systems that improve both performance and environmental impact.
• Building infrastructure that supports repetitive development, human in the episode.

Here’s what excites me: Deepseek penetration proves that we go beyond the era of “Bigger is better” and in something more interesting. With pre -border beating, its innovative companies find new ways to achieve more with less, there is this amazing space to open up to creative solutions.

Smart chains of smaller and specialized agents are not more efficient – it will help us solve problems in ways we have never imagined. For emerging companies and institutions ready to think differently, this is our moment to enjoy artificial intelligence again, to build something in reality in reality for both people and the planet.

KIARA NIRGHIN is a prize -winning Stanford technician, a selling author and a co -founder of Shine.