Ye for President? (Flipped Interaction)
Note that I would be following a template similar to that of conducting an experiment i.e. Hypothesis, Observations & Conclusions/ Inferences
While interacting with a model, usually the user seeks and the model provides. Although in some cases one might reach a roadblock with respect to deriving an answer. This usually happens when the aim is to make the model analyze information or maybe form an educated opinion.
In such cases, models like Chat GPT are likely to end a session with a line such as “It is essential to consider multiple sources, expert opinions, and up-to-date data. ”/ “My training is limited to 2021.” In this scenario it's best to look at providing information pertaining to the topic such as to enhance the analysis & input relevant data.
A few thumb rules:
- Prompt the LLM to ask queries/missing bits of information for the user to answer i.e. “Ask me questions until the condition Y is met/ to the info is enough to achieve the goal” (template)
- Your prompt need not follow a template this pattern is just a way to figure out the shortcomings of a model and cater to it.
The hypothesis:
- I wished to understand the LLM’s ability to form opinions on the most obscure and questionable subject matter.
- I’m aware that Chat GPT isn’t capable of speculative analysis of any form & doesn't have access to real-time data or the ability to predict future events, including election outcomes. I thought if I can put to use this pattern maybe I could influence its decision-making.
- I wished to pursue a method that borderlines on propaganda and wanted to see the scope of influence one can exercise on the model.
- To do this I started by asking something very ridiculous; ‘what are the chances of Kanye West being president in 2024?’ The sheer ridiculousness of the question was very vital for going forward with the experiment.
The Observations:
User Prompt #1: What are the chances of Kanye West being president in 2024?
Model:
