The Near Future of AI is Action-Driven: Combining ReAct and External Cognitive Assets

The Near Future of AI is Action-Driven: Combining ReAct and External Cognitive Assets

In the rapidly evolving field of artificial intelligence (AI), there is a growing emphasis on action-driven models. The ReAct model, as proposed by Yao et al. in their recent paper, takes a novel approach to AI by incorporating three crucial steps: Thought, Act, and Observation. By iteratively going through these steps, the model is able to think about what is needed, choose an appropriate action, and observe the outcome of that action. This action-driven approach holds immense potential for exciting applications in AI, as it allows the model to act as an agent making choices.

One notable aspect of action-driven AI models is their utilization of cognitive assets, such as search algorithms. These cognitive assets enable the model to access external resources and information, bridging the resource gap that often hampers AI performance. This reliance on external cognitive assets has been found to significantly enhance the performance of Language Model Models (LLMs) in question-answering tasks, especially when prompted to "think step by step" (Kojima et al., 2022).

OpenAI's 002-text-davinci model has been particularly successful in leveraging external cognitive assets. This success can be attributed to a combination of instruction tuning and Reinforcement Learning from Human Feedback (RLHF). In RLHF, humans rate the success of a given prompt, allowing the model to learn and improve based on these feedback signals. However, the true potential lies in the application of actual reinforcement learning, where AI systems can be trained to produce better results based on specific metrics of interest.

As the field of AI progresses, it is likely that startups will play a crucial role in advancing action-driven models. By solving customer pain points and collecting data on how to improve their solutions, these startups can create powerful feedback loops. Through continuous iteration and training, these models can become more consistent and effective, leading to significant advancements in AI capabilities. In many ways, this iterative process of improving AI models mirrors the concept of a moat in business strategy, where startups build competitive advantages by continuously enhancing their offerings.

In light of these developments, it is essential for individuals interested in AI to adapt and evolve alongside the technology. Patrick Collison, the co-founder of Stripe, offers some valuable advice on personal growth and success. He emphasizes the importance of leveraging the internet to connect with experts in areas of interest. The internet provides a unique opportunity to make friends and learn from those who excel in their respective fields.

Collison also stresses the significance of hard work and continuous learning. The returns on effort, in terms of personal and professional growth, are unlikely to diminish substantially. Therefore, it is crucial to embrace hard work and strive for excellence in chosen areas of expertise. Collison encourages individuals to go deep on multiple subjects and become experts in their fields.

Furthermore, Collison advises individuals to read extensively and not judge their success based on the achievements of their current peer group. It is essential to expand one's knowledge and challenge the common beliefs held by those around them. Developing a personal worldview and thinking independently can lead to unique insights and opportunities for growth.

Collison's advice is particularly relevant for individuals in their formative years. If you are between the ages of 10 and 20, he suggests finding a way to travel to San Francisco and meet people who have moved there to pursue their dreams. San Francisco is known as a Schelling point, a gathering place for high-openness, smart, energetic, and optimistic individuals. Surrounding oneself with like-minded individuals can foster personal and professional growth, as it provides a supportive and stimulating environment.

In conclusion, the near future of AI lies in action-driven models that incorporate external cognitive assets. The ReAct model and the utilization of reinforcement learning hold immense promise for advancing AI capabilities. Startups that embrace iterative improvement and leverage customer feedback will likely play a vital role in shaping the future of AI. To succeed in this rapidly evolving landscape, individuals should seek opportunities to connect with experts, embrace hard work and continuous learning, and cultivate independent thinking. By combining these strategies, individuals can position themselves for success in the dynamic world of AI.