Want to get more insight into AI developments? Then this extensive interview "What if Dario Amodei is right about AI?" between journalist Ezra Klein of The New York Times and Dario Amodei, the founder and CEO of Anthropic, is highly recommended.  

Anthropic is the company behind the leading AI model Claude, a competitor to ChatGPT. Unfortunately, only the limited free version of Claude is available in the Netherlands, not the more powerful Pro version that offers many more features.

When Dario Amodei was still at OpenAI (between 2016-2020), where he was vice president of research, he began to wonder what would happen if you fed AI more and more data. Together with his colleagues, he researched this. They found that AI not only performed better with more data, but that this improvement was exponential. The growth of AI capabilities started slowly at first, but then skyrocketed. This growth curve showed the characteristic shape of a field hockey stick.

Today, Amodei is the CEO of AI company Anthropic. He still believes that AI is growing exponentially. This is based on so-called scaling laws. He thinks we are now in the steep rise of that growth curve.

One of Amodei's predictions is about the development of AI models in the next three months to two years. He believes that a new generation of language models will become available every 4 to 8 months and that AI can start performing complex tasks with 99.9% accuracy. Such as writing code and running tests. These AI agents will be able to perform sequences of 20 to 30 steps accurately, according to Amodei. This is an important step for practical AI applications for multi-step work, as is common in marketing, among others. 

Amodei also predicts that AI models will soon have more personality, while remaining objective and actionable. This could greatly accelerate the adoption of AI, as users will feel more connected to this technology.

Enabling these developments and staying ahead requires massive investments that only a handful of large tech companies such as Microsoft, Google, Meta, Apple and Amazon can afford. According to Amodei, training the current language models costs about $100 million. The new generation of models coming out later this year will easily require $1 billion. And investments run as high as $5 to $10 billion for the models expected in 2025 and 2026! 

The interview was not only about the possibilities of AI, but also the ethical and security challenges. It is crucial that AI models can understand "truth" and ethics. Especially in a world where people do not always agree on what is true.  

This talk highlights the enormous developments and potential of AI, but also makes clear the importance of being prepared for the high rate of change that is predicted.

Want to exchange thoughts on how to implement AI within your marketing team? Let us know, we'd love to help.