The Future of Technology: From Innovation to Action-Driven AI

The Future of Technology: From Innovation to Action-Driven AI

In today's rapidly advancing technological landscape, it is crucial for companies and innovators to not only focus on invention and innovation but also on the ability to commercialize their ideas and bring them to the masses. Larry Page, the co-founder of Google, emphasizes the importance of combining both aspects to create a positive impact on the world and give people hope.

Page believes that companies should strive for revolutionary change rather than incremental change, especially in the field of technology. He suggests that many companies fail over time because they miss out on the future. To avoid this, he encourages organizations to constantly look ahead and create the future they envision.

One key concept that Page mentions is "additionality," which refers to doing something that wouldn't happen unless you were actually doing it. By focusing on areas that others might not consider or working on projects that no one else is working on, companies can achieve true additionality and make a significant impact.

In the realm of artificial intelligence (AI), the near future is action-driven. The ReAct model, developed by Yao et al., emphasizes the importance of thought, action, and observation in an iterative process. By choosing actions and observing their outcomes, AI models can act as agents and make decisions that drive progress. This action-driven approach aligns closely with the idea of additionality, where AI systems can go beyond what is expected and deliver revolutionary change.

For AI models like language learning models (LLMs), the ability to think step by step and utilize external cognitive assets is crucial. Kojima et al. found that LLMs perform better at question-answering tasks when prompted to think step by step. However, their performance can be further enhanced by accessing external resources or cognitive assets. By fetching data from external spaces, LLMs can bridge the resource gap and achieve even better results.

OpenAI's 002-text-davinci model has demonstrated remarkable performance, thanks to a combination of instruction tuning and reinforcement learning from human feedback (RLHF). The model's success lies in its ability to learn and improve based on human ratings of prompt success. However, the true potential of AI lies in actual reinforcement learning, where systems can be trained to produce better results through continuous improvement.

In the AI startup ecosystem, successful companies will create powerful feedback loops by solving customer pain points, collecting data to enhance their models, and iterating on their solutions. This iterative process allows them to bootstrap their way to success and develop a competitive advantage in the AI landscape. This approach forms the foundation of what a moat will look like in AI for the foreseeable future.