The author asserts that while LLMs are revolutionary and can automate mundane tasks, they are not intelligent by the first principles of intelligence. The hype around LLMs leading to AGI is attributed to humans equating language with intelligence. The author believes that achieving AGI will require new architectures beyond current software and hardware capabilities, and the ability to develop a mental model of the world. For now, LLMs and their genAI Copilots are seen as Assistive Intelligence.
Key takeaways:
- The concept of a programming Copilot, introduced by Fred Brooks, has evolved from a human alter-ego to a genAI coding assistant, assisting in various domains beyond programming.
- Large Language Models (LLMs), the technology behind genAI copilots, are not true AI and are not close to achieving Artificial General Intelligence (AGI). They are essentially models that predict the next token in the output they are generating, without understanding the content.
- LLMs are revolutionary and have the potential to take over mundane, repetitive, or error-prone human tasks, but they are not capable of becoming AGI as they are not designed to operate outside of their programmed lanes.
- AGI is an achievable ambition but will require new architectures beyond current technology, including the development of a mental model of the world the AGI will need to operate autonomously in.