AI's Big Bet: Is Bigger Always Better?

Big tech companies are spending a lot of money to build huge AI systems. They believe that making these systems bigger will make them smarter. But not everyone agrees with this idea. Some experts think that just making AI models bigger is not the best way to improve them. They believe that AI needs to learn from real-world experiences, not just from lots of data. One of these experts is Sara Hooker. She used to work at Cohere and Google. Now, she has started her own company, Adaption Labs. She wants to build AI systems that can learn and adapt on their own. Hooker says that current AI systems are not very good at learning from their mistakes. For example, if an AI system makes a mistake, it does not learn from it. It just keeps making the same mistake. Some companies offer to help businesses adjust their AI systems to their specific needs. But this can be very expensive. OpenAI, for example, reportedly charges millions of dollars for this service. Hooker believes that AI systems should be able to learn from their environment. She thinks that this would make AI more accessible and affordable. Adaption Labs is not the only company that is questioning the idea of scaling AI models. Some researchers have found that bigger AI models may not always be better. They have found that the benefits of making AI models bigger are decreasing. Other researchers are exploring new ways to improve AI. For example, some researchers are working on AI reasoning models. These models take more time and resources to work through problems. But they can push the capabilities of AI models even further. Adaption Labs wants to find a new breakthrough in AI. They want to prove that learning from experience can be cheaper and more effective than just making AI models bigger. Hooker has experience in building small AI models. She has shown that these models can outperform bigger ones in some tasks. She wants to continue this trend. Hooker also believes in making AI research more accessible. She has hired researchers from all over the world, including underrepresented regions like Africa. She plans to continue this approach with Adaption Labs. If Hooker and Adaption Labs are right, the implications could be huge. It could change the way we think about AI and its potential.