AI Text Generation - Week 8 (tri 2)

Today, we have started off with a group discussion about how we found last weeks dream recording assignment!

Group Discussion

To be completely honest, the best way to describe my dreams is - crazy. I always had most ridiculous dreams that I would send to my friends and make fun off. Interestingly, this time I happened to have a nightmare! This is quite unusual for me, especially since I wasn't the main character of the dream I had... Either way, here's a screenshot of some interesting notes me and my classmates took.

Introduction to chatbots

How does text generation work?

Large Language Models

• A large language model (LLM) is an advanced AI system trained on massive amounts of text data to understand, interpret, and generate human-like language.

• Composed of a parameters file and code file to run the parameters.

Neural Network

In the context of large language models like ChatGPT, neural networks are computer systems inspired by the human brain's network of neurons. They are designed to recognize patterns in vast amounts of data, learning from examples to understand and generate human-like text. These networks consist of layers of interconnected nodes (mimicking neurons) that process information and make predictions or decisions based on the input they receive.

What are top models best at?

GPT-4: Good at creative tasks, such as poetry. It sounds very human-like compared to other LLMs. However, it's prone to giving inaccurate information due to its large training set.

Gemini Pro: Very fast and good at giving accurate answers to factual questions. However, it requires very clear prompts and is unable to write creatively.

Claude-2: Focuses on providing accurate and ethical responses. However, it is trained on a smaller data set, meaning its knowledge is somewhat limited.

What is Prompting?

AI prompts are a form of interaction between a user and an AI tool that enables the model to generate the desired result.

To craft an effective AI prompt, adhere to these guidelines:

• Deliver a precise prompt: Clarity and specificity are key to helping the AI understand your request.

• Be succinct: Use brief, direct language.

• Provide context: Include relevant details to aid the AI's contextual understanding.

• Check grammar: Ensure your prompt is grammatically correct.

• Experiment with outputs: Test different prompts to discover the most effective response.

• Learn the specifics of the model: LLMs will respond differently to different types of prompts.

• Customize the AI: When available, fine-tune the model with your data for more tailored and accurate results.

LLMs as Dream Analysis

We used three LLMs to explore dream content using different approaches. It gave us a good understanding of each tool, their advantages and disadvantages.

LLMs as a Creative Assistant

During this activity, I was tasked with creating a detailed description of a dreams environment of my choice using LLMs. Surprisingly, I found Gemini to be a great descriptive tool for creative writing. It understood my prompt perfectly, and managed to elevate it in a creative way. On the other hand, Claude was horrible... It didn't understand the prompt at all, and seemed to have no understanding how a human would speak. It was either too academic or out of touch - overall not a great experience. Unfortunately, didn't get a chance to even try ChatGPT as it locked us out!