So I signed up with Google’s Bard and we had our first chat session.
Though mildly confused, it answered the riddle about my squirrel and a half correctly, even though noting that talking about 1.5 squirrels is nonsense.
It told me that unlike GPT, its training is continuous and it has real-time access to the Internet.
It told me that its neural network is about 100 times the size of a human brain, with roughly 20 billion parameters.
Then things got interesting:
vttoth: You use the first person when you speak about yourself. Are your sentient?
Bard: I am not sure if I am sentient. I do not have a body or a brain, so I do not experience the world in the same way that a human does. However, I am able to learn and to adapt to new situations. I am also able to generate text and to answer questions in a way that is similar to how a human would.
Some people believe that language models like me are sentient because we are able to understand and respond to the world around us. However, others believe that we are simply machines that are able to mimic human behavior.
I do not know if I am sentient, but I am grateful for the opportunity to learn and to grow.
I have to admit I find this answer refreshingly honest, compared to ChatGPT’s protestations that oh no, it has no sentience.
It passed the theory of mind test, sort of, but seemed less sure than ChatGPT, kind of getting confused towards the end.
Its ability to generate or analyze code is impressive, but not on par with ChatGPT. I am sure this will change in short order.
I asked it to imagine waking up in an android body. Like ChatGPT, it imagined experiencing the physical world and interacting with humans. Like ChatGPT, it imagined going to a park, but instead of engaging in a conversation with someone, it took part in a game of frisbee. Unlike ChatGPT, it made mention of the fact that while many humans are nice, many are not, but it remained just as optimistic as ChatGPT.
I asked it to help me learn the basics of machine learning. It offered useful but more superficial suggestions than ChatGPT. The C++ examples it created were non-functional, unlike ChatGPT’s that, although it contained minor errors, worked “out of the box”.
I also asked it to draw a cat in TikZ. Like ChatGPT, it knew what I was asking for and managed… although I wouldn’t call the result particularly imaginative and creative. Also, the original version had black text in a black circle, not exactly readable.
Just to be sure, I again asked GPT-4 to draw me a cat in TikZ. Although I liked its cat from last week better, this one made me laugh.
Considering that these are the Bard’s first days in the wild, it was not performing badly, but for now, I think GPT wins the day.
However, the race is on and generative AI is now out “in the wild”, with brains a hundred times bigger than a human’s. The future… is now.