Paper short abstract:

What happens when CHATGPT API access becomes more closed through charging or blocking certain functions? How much do the results of a same function change when the prompts varied?

Paper long abstract:

CHATGPT uses neural networks to train and feed models that are then introduced into the market, but we do not know how they work inside. These neural networks obscure the way OPENAI (CHATGPT's company) designs its models, as well as its protocols for other applications to connect. The price changes depending on the chosen version, which also causes the model to vary in capacity or access. To communicate with the CHATGPT programming interface, it is necessary to resort to its API. Following Jünger, the Social Sciences, Economics, and Engineering have been using this type of APIs not so much to analyze as to collect data. What happens when access for this collection becomes more closed through charging or blocking certain functions? How much do the results of the same function change when the prompts are varied? This study proposes an explanatory mixed methodology. It begins with an experimental quantitative approach using a multi-section code to analyze feelings that can modify variables such as the model, temperature, and prompt. By varying the way of requesting processing, the results change. It is assumed that it should not do so in that way. This is where a qualitative analysis from linguistics and Science and Technology Studies comes in, responding to the experimental variations of the request and the results.