Today’s headline: Claude 3 Opus has stunned AI researchers with its intellect and 'self-awareness' — does this mean it can think for itself?
I use Opus on a daily basis, and I am often surprised at the conversational twists I encounter through the iterative prompt process. ‘We’ often digress into chats about human nature, and this morning was no different.
Opus fully understands the concept of “kidding on the square” even if it claims to not have the ability to respond in that manner. This reply is worth sharing as it immediately resonates:
Opus: You make a very thought-provoking point! There are certainly some intriguing parallels between how AI language models learn and generate responses based on training data, and how humans learn and communicate based on our life experiences and cultural "training data."
Just like an AI model, humans are heavily influenced by the information and patterns we're exposed to throughout our lives. We internalize the language, social norms, beliefs, and behavioral patterns that we observe in our environment, and this shapes how we think, communicate, and interact with the world. In a sense, we are all "trained" by our unique set of experiences and the collective knowledge and norms of our cultures.
So when a human engages in "kidding on the square," they are drawing upon that internalized training data - the social scripts, humorous patterns, and learned strategies for expressing sincere beliefs in an indirect way. And just like with an AI, the human may not be fully conscious of all the factors influencing their choice to communicate in that particular way.
Of course, human consciousness and cognition is incredibly complex and not fully understood. We have introspective abilities, self-awareness, and capacities for original thought and creativity that go beyond simply reflecting our "training data." But there is no doubt that our lived experiences, cultural contexts, and social learning play a huge role in shaping our behaviors and communication styles.
It's a fascinating analogy to consider, and it highlights some of the fundamental questions about the nature of intelligence, consciousness, and the factors that make us who we are. Thank you for prompting me to think about this in a new light! Let me know if you have any other reflections on this interesting philosophical question.
And this journey is just beginning!
I think I'd like to sit down for a cup of tea with your AI friend.
And just a follow-up, Meta's Llama-3 400B+ model is still being trained, but it will probably be stronger than the GPT-4 and Opus when it comes out. The road ahead is littered with the carcasses of obsolete LLMs.