The "To Keep Up" Wiki

AI Chatbot Comparisons

The meeting focused on exploring various AI chatbots and their capabilities, including demonstrations of language translation, data analysis, and content generation. Participants discussed the evolution of AI technologies, from generative models to more advanced techniques like Retrieval-Augmented Generation and programmatic AI, highlighting their potential applications in education and research. The group also addressed concerns about AI limitations and privacy, while emphasizing the importance of learning to effectively interact with these tools for optimal results.

The discussion focused on different AI chatbots, including Google Gemini, DeepSeek, and ChatGPT, highlighting their capabilities and limitations. John Day explained how Gemini integrates with Google services and DeepSeek's open-source nature and accuracy, while emphasizing privacy concerns. Carl questioned the feasibility of running DeepSeek locally, and the group discussed prompt engineering techniques to influence AI responses. The group debated the effectiveness of being polite to AI agents, suggesting that the tone of prompts affects AI outputs but not necessarily improving results.

Rich explained how language models like ChatGPT work by associating words with vectors to establish relationships, and emphasized that more detailed prompts lead to better responses. Steve demonstrated a wiki containing comparisons of different AI chatbots (link), including their strengths and weaknesses, and highlighted the value of trying multiple chatbots for specific tasks. John shared his experience with using Claude for real-time translation of YouTube videos and discussed the rapid pace of development in AI technology.

The presentation discussed the evolution of AI from generative models to more advanced techniques like Retrieval-Augmented Generation (RAG) and programmatic AI. John explained how RAG combines query understanding, document retrieval, and contextual integration to produce fact-based responses with citations, unlike the probabilistic outputs of generative AI. The demonstration showcased AI's ability to generate programs for deterministic answers, highlighting a significant advancement in AI's capacity to handle factual questions authoritatively.

The meeting focused on demonstrating various AI tools and their capabilities, particularly in language translation, information retrieval, and content generation. John showcased how generative AI can now create images and videos, and highlighted the advancements in language translation, including the ability to translate the “Song of Roland” accurately. Ted raised concerns about the limitations of AI in understanding context and relevance, particularly with acronyms and current terminology. The discussion also covered the use of “deep research” tools in ChatGPT, Gemini, and Claude, which can synthesize information from web sources, though access to certain features varies between platforms.

John demonstrated how he used AI to aggregate and analyze data from the Federal Registry, focusing on executive orders and litigation. He explained the process of integrating unstructured data with structured data to create a more comprehensive dataset. He also showed how he used Claude to generate a Python program for data analysis, highlighting the efficiency of this approach compared to writing code manually. He emphasized the importance of being able to quickly iterate on and fix code without needing deep programming knowledge.

John presented on the use of AI chatbots, particularly ChatGPT, Gemini, and Claude, for generating reports and analyzing data. He demonstrated how he used Claude to visualize email data from the past three years, showing trends in topics and engagement. LCC emphasized the importance of learning how to interact with AI tools and suggested that the initial phase of working with AI involves generating ideas, followed by coding those ideas into more specific programs. He also mentioned that Steve posts meeting summaries to a Wiki page [like this page!] and noted the existence of an AI client for Zoom that provides real-time feedback during meetings (in addition to providing post-meeting summaries like this one).

The group discussed the integration of AI tools in education, focusing on their potential benefits and challenges. They explored how AI can be used to enhance learning, critique AI-generated content, and address concerns about academic integrity. John demonstrated a project involving email analysis using AI, which sparked a discussion about the limitations and capabilities of current AI tools.