Chat Model Showdown: Part 2 — AI Generated Video of the Debate

I’ve taken a slight deviation from my initial plan, and Part 2 is landing sooner than expected! If you haven’t checked out Part 1, I encourage you to read it <a href="https://medium.com/@hominum_universalis/chat-model-showdown-gpt4-vs-gpt-3-5-vs-llama-2-in-a-simulated-debate-part-1-41e4f8dcc8bd" rel="noopener">here</a>. To recap Part 1, I formulated a debate simulation between GPT-4, GPT-3.5, and LlaMA2 using Python and LangChain. My main goal was exploratory in nature, aimed at gaining insight into how well chat models can simulate reasoning. The discussion of the simulation results focused on the chat model’s discourse regarding the proposition: “Should AI be granted Legal Personhood?” To explore advancements in generative AI outside of basic text generation, I transformed the debate’s arguments and the judge’s evaluations into a video, using <a href="https://www.synthesia.io/" rel="noopener ugc nofollow" target="_blank">Synthesia</a> and <a href="https://www.midjourney.com/" rel="noopener ugc nofollow" target="_blank">MidJourney</a>. While this process wasn’t fully automated, it’s not hard to imagine a future where this becomes a seamless, programmatic procedure. The end result was genuinely impressive. Viewing the debate in video format brings a different but very important aspect to our evaluation of how effectively these models can mimic reasoning. Dive in an see for yourself. <a href="https://medium.com/@hominum_universalis/chat-model-showdown-gpt-4-vs-a60cbce5f0ce">Read More</a>