In our conversation yesterday, we touched on how each new Large Language Model (LLM) gets branded as the next "GPT killer." But I think aiming to outdo OpenAI and its GPT series isn't really where industry's focus should be. It's more about broadening what Generative AI can do and opening up new avenues. In a parallel narrative, ChatGPT and other conversational LLMs are increasingly being recognized as potential challengers to Google's long-standing dominance in the search arena.
"War is peace, freedom is slavery, ignorance is strength" - George Orwell, "1984"
The predicament surrounding platforms like ChatGPT stems from their seemingly vast knowledge, which paradoxically can be seen as a double-edged sword. With knowledge comes the expectation to adhere to high moral standards and make ethically sound choices. On the other hand, Google basically works as a navigator, guiding people to various websites. This role as a navigator keeps it away from moral debates, unlike platforms that offer direct knowledge.
The high moral expectations placed on OpenAI and similar entities don't guarantee human compliance especially when moral standards are subject to convenience. The solution might be in developing AI models that are both localized and personalized, tailored to align with the unique tastes and preferences of users. This approach could foster a digital environment where individuals engage with content that resonates with them, regardless of broader moral standards.
Imagine a future where customizing a Large Language Model (LLM) for your desktop is as easy as downloading your favorite app on your smartphone. In this envisioned future, a dynamic marketplace thrives, with specialized agents that tailor your LLM to match your unique tastes and needs. This personalization, crucial for creating a more relevant and engaging AI experience, primarily involves fine-tuning the model's weights to mirror your specific preferences – a possibility exclusive to open-weight customizable LLMs.
Transitioning from customization to functionality, these LLMs in their 'generative agents' form – a concept distinct from 'foundation agents' involving robotics – excel in two crucial areas: possessing extensive knowledge and performing tasks. The link between these two capabilities is the decision-making capacity, or agency. While achieving complete agency will be a step-by-step process, it initially involves a synergistic relationship between humans and AI, combining our insights with their analytical power.
The capability to successively enhance these LLMs unveils a world of fascinating opportunities. Imagine a film enthusiast accessing uncut, director’s commentary versions of movies — content often overlooked by mainstream platforms. Or consider a music producer delving into underground genres and rare samples, realms where general AI recommendations might not tread. Equally, picture a street art aficionado exploring global artworks, including those in legal gray zones, typically bypassed by standard AI models due to copyright nuances.
While it's imperative that no LLM ever allows what is clearly immoral, such as actions harmful to humanity, there's a vast landscape of moral gray areas to consider. It's within these nuanced spaces that localized, personalized LLMs can truly excel, adeptly navigating complexities that standard models might overlook. This shift in AI, from a simple tool to a perceptive companion, heralds a future where each individual's unique journey of discovery and understanding is not just recognized but intimately supported.