AI 101 Live Session, 2024-02-23


Recorded live AI Salon / AI 101 session on AI topics.

Next: AI 101, Let's Learn About APIs (and a tiny bit of trying Claude API), 2024-02-24

Previous: AI 101 Live Session, 2024-02-22

AI Summary (hopefully useful, may be inaccurate)

Quick Recap

The discussion revolved around the limitations and potential of language learning models (LLMs) and artificial intelligence (AI) in teaching and testing. The team also considered the development of a personalized learning curriculum and the concept of a school operating like a Wiki. They explored the integration of LLMs into their teaching, the creation of an AI glossary for personal use, and the application of computational thinking to learning and teaching. The conversation also touched upon the development of a generative AI vocabulary project and the use of GPT to simplify and slow down processes.


Limits of Language Learning Models and AI's Potential in Personal Life

The discussion centered on the limitations of Language Learning Models (Llms) in teaching and testing topics. The speaker expressed uncertainty about the effectiveness of Llms in these roles, despite their ability to overstate information. The speaker also suggested that Llms could be trained to teach and test, but the effectiveness of such a system was uncertain. Pete and R then shifted the conversation to the potential uses of artificial intelligence (AI) in personal life. R expressed interest in AI's ability to learn, teach, and test subjects in-depth. Pete summarized R's goals, emphasizing the potential of AI as a powerful tool for learning and teaching.

LLMs for High-Level Teaching and Customized Instruction

Scott proposed the use of a low-cost method, such as an LLM, to provide high-level teaching. R added the idea of customized instruction and the potential for LLMs to create programmed learning curriculums. Pete agreed with R's suggestion that LLMs could be used to teach a person or set up a structure to help teach a subject. They also discussed the limitations and potential of AI systems, particularly in the context of module organization and comprehension, and agreed that a comprehensive system could be developed using existing content and a simple, reliable approach.

AI Glossary and Personalized Learning Curriculum Development

The team discussed the creation of an AI glossary for personal use, with a focus on beginners. Julia expressed interest in using it for a school project. They agreed to rank the content, avoid technical language, and include a disclaimer that the glossary is under development. The team also discussed the development of a personalized learning curriculum, emphasizing the importance of knowing a learner's resources and teaching style. They planned to divide the curriculum into sections for beginners and more advanced learners, and test methods such as asking AI for suggestions. The team also explored the concept of a school operating like a Wiki, allowing students to learn independently with the support of facilitators, and the potential of AI in education, such as using chatbots as teaching aids.

Incorporating Learning Language Models in Teaching

Pete and R discussed incorporating Learning Language Models into their teaching. They considered having students complete exercises, with Scott possibly providing a demo, and Pete suggested using jailbreaking as a real-world application. They also discussed OpenAI's ChatGPT, with Pete expressing concerns over lack of documentation and intellectual property theft, while R suggested staying on topic. They ended by discussing the concept of computational thinking, with Pete suggesting a homework assignment to ask ChatGPT about it and the possibility of jailbreaking ChatGPT for additional functionality.

Computational Thinking in Learning and Teaching

Pete discussed the application of computational thinking to learning and teaching a topic, using the example of World War One. He expressed a need for the system to generalize across topics, rather than focusing on a specific one. Pete suggested that the system could be used to generate lists or exercises on a particular topic, but would need human setup to apply them. He also emphasized that the system should be structured in a modular way, with a human to oversee the overall framework. Pete also mentioned the possibility of using an LLM for teaching and testing.

Slowing Down and Simplifying Project

Pete and R discussed the need to slow down and simplify their project, proposing the idea of opening another window to control the conversation pace. They explored the challenges of building an effective teaching system, emphasizing the importance of focusing on relevant information and structuring prompts. The discussion also touched upon the idea of a GPT that could help users create effective prompts using structured frameworks. The participants suggested that learning how to structure a prompt properly could save time and improve the learning process. The possibility of using LLMs for testing understandings and providing customized learning was also discussed. Towards the end, they considered building a small pedagogical system to help users learn at their own pace, and the potential use of Markdown to enhance the user experience was briefly discussed.

AI Vocabulary Project Development

Pete and R discussed the development of a generative AI vocabulary project. They decided to start with a beginner level approach, focusing on the right words rather than quantity. Pete suggested creating a dictionary of AI terminology from scratch to save time. They also considered the use of hard-coded instructions and the idea of creating a machine learning AI. The conversation ended with Pete expressing interest in continuing the project next week.

GPT Vocabulary Tutor for Tech-Illiterate Users

Scott discussed his experiences and findings with a GPT creator. He shared that he had asked the GPT to create a basic vocabulary tutor for a tech-illiterate person, like his mother, and was impressed with the results. He suggested that this method could be an effective way to introduce basic concepts to those who are not tech-savvy. However, he also expressed some confusion about how to share the prompts he had used with the GPT, as they seemed to disappear when he edited them.

AI Simplification Experiment Discussion

Scott and Anita had a conversation about using AI, specifically Gpt, to simplify and slow down processes. Scott suggested relying on Gpt to understand needs rather than the other way around. Anita proposed a different approach, such as breaking down technical concepts to the elementary level. They agreed to experiment with these ideas to see if they yielded improvements.