In this meeting, the State Changers discussed the following topics:
The main focus of the discussion was on using APIs, specifically with OpenAI's ChatGPT, and trying to get the desired content delivered in a JSON format. They explored ways to structure the input text clearer to get the desired output. They also discussed issues related to variable outputs and inconsistency in data structures. One solution they proposed was to set up a conversational loop with ChatGPT that could return an error if the output was not as desired. They also discussed adjusting and testing the input text within the OpenAI playground for faster learning and results. The State Changers also explored potential error handling in scenarios where the API gives no results or an error message and discussed the possibility of using retry loops to address these issues. Another fascinating topic they discussed was the inherent nondeterminism within machine learning technologies. They emphasized the importance of handling possible variations and errors that might arise from this stochastic nature, and how this can also lead to certain advantages, like the ability to generate different answers each time. To conclude, they mentioned making the code and the tech stacks more complicated to significantly improve reliability, despite the inherent challenges with this approach. As per the transcript, the following mentioned keywords were discussed: "JavaScript", "OpenAI", "AI21", "JSON", "State Change", and "Xano".
(Source: Office Hours 5/10 )
