The meeting among the State Changers involved discussions about coding problem related to color editing and new chart creation using a script tag. One participant had trouble implementing changes in code, and these were resolved during the meeting.
A significant portion was dedicated to discussing AI's capabilities and limitations, particularly the OpenAI's GPT-3. The tool's memory and reasoning power were lauded, but it was acknowledged that its comprehension of world knowledge was "stump-like". To improve this, they discussed the process of document retrieval to pull relevant data into the context of the AI prompt. The distinction was made between the three levels of understanding; showing up for an exam without studying (comparison for ChatGPT's lack of in-depth knowledge), preparing for the exam beforehand, akin to fine-tuning a model's memory, and having an open-book exam, where the information is right in front of you, as suggested for upcoming AI improvements. Lastly, the discussion headed towards understanding how an AI can be trained, using a stream API, for different data like a book or personality, as seen on a site by company Delphi. They did however forecast that the newfound ways of monetizing the AI API might lead to market saturation in a few years. The importance of a good interface to derive value was highlighted. They concluded by encouraging further exploration of Chart.JS, an open-source JavaScript library for data visualization, but with a healthy apprehension belied by some issues discussed over the meeting. The participant was encouraged to attend the following office hours for any further code troubleshooting needs.
(Source: Office Hours 9/20/2023 )
