This research diary is where I am going to record my progress and struggles in my AI research at my university. Hopefully the contents of this blog help me reflect on my growth in the future, showing me how far I have come since day one!

For now my focus is on studying how AI is made and the big names in LLMs (Large Language Models). I am centering my research on the most popular ones, such as BERT and the various GPT models, finding what they are used for and how they work (as far as we know)

So far there have been a couple of things that interest me. First, I find it very interesting that no one knows exactly how LLMs function. We know that they use word vectors but how exactly they are used seems to be a bit of a mystery as far as my research goes. LLMs do things their own way and I find that fascinating. I wonder if we can ever get them to explain it to us in terms we can understand!

The other thing I find interesting is a bit of a game. So far, every model I’ve been able to use to some extent I have asked a simple question: how many “r”s are there in the word “strawberry”? To humans that can read and count the answer is obvious: 3. But to LLMs it doesn’t appear to be that straightforward. All of the models I’ve been able to ask, GPT-4, GPT-4o, Claude, and Copilot all at first answer 2. They can be coached to the correct answer but I think it is interesting that they all give the same incorrect answer! I will definitely be continuing this little game as I research.