Week 1

So for the first week, I started learning the mathematics behind machine learning and data science. I also attended an event, Generative AI in Action Roadshow, which also was very helpful.

ML Mathematics

I'm taking this course on Coursera by Imperial College London. At first it was easy since I learnt vectors in school, but then it got a bit harder to grasp the concept and how it applies to machine learning. It's only the first week so I'll probably understand further on. To be specific I learnt about the dot product of vectors, vector projection, linear dependency and how to change basis vectors. To oversimplify it, I guess data can be structured into vectors which then can be used to predict how to get the desired outcome using vector operations?

Generative AI Roadshow

During the GenAI event, I learnt what is generative AI and how it works. From what I understand, it's a newer form of AI that takes a prompt and generates (hence "generative") new things such as images, video, code, etc. Then I learnt that LLMs just have weights, and AI agents are needed ( which are basically the model(s) (the brain), the tools (arms & legs) and how it's connected with goals, reasoning, etc. ) to do things

There's a no code/low code solution to making generative AI on Google Cloud which has multiple services, one of them being Vertex AI which has an agent builder as well. Even though uploading data there took quite a long time, it's actually good for businesses that want to implement generative AI in their websites or other things.

For a full code method of making generative AI, we learnt function calling which is a way for AI to retrieve more relevant or real-time information instead of using RAG to give set data. How I think of it, is that functions are used to retrieve data from an API, then create a more human response from the data. Something interesting that I learnt was a syntax that implies calling a function without hard coding it to make it more flexible. Something about a double pointer, I'm going to have to research more about that.

# assign function name and arguments as provided in the function call response
func_name = response.candidates[0].content.parts[0].function_call.name
func_args = response.candidates[0].content.parts[0].function_call.args

# call the function to get the exchange rate information
fx_rate_json = globals()[func_name](**func_args)
fx_rate_json

There was also some things about Crew AI about using multiple agents to achieve a certain goal.

Additional Stuff

I also did more research and started reading some books. Now I know some basics of neural networks, such as how it "learns" using costs and weights to create a gradient descent. These videos by 3Blue1Brown was very helpful.

Wrap Up

That's all for this week, for those reading, please let me know if I got anything wrong! I might have left some things out but it's probably enough for future me to review.