The world of AI chatbots is a whirlwind of innovation, with new developments and surprises seemingly emerging every week! Since the end of April, one particular model, modestly gpt2-chatbot, captured the attention of myself and other AI enthusiasts due to its advanced abilities and sparked much speculation. This mysterious bot first appeared on April 28, 2024 on LMSYS Chatbot Arena, vanished two day later, and has now resurfaced on the LMSYS Chatbot Arena (battle) tab, ready to compete against other AI models. Its sudden appearance and impressive capabilities have left many wondering about its origins and potential, with some even theorizing it could be a glimpse into the future of AI language models.

The Mystery of gpt2-chatbot

Beginning on April 28, chatter about a new gpt2-chatbot started circulating on the internetz, with experts expressing both excitement and bewilderment over its advanced capabilities. The model, which appeared without fanfare on a popular AI testing website, has demonstrated performance that matches and potentially exceeds that of GPT-4, the most advanced system unveiled by OpenAI to date. Researchers like Andrew Gao and Ethan Mollick have noted gpt2-chatbot’s impressive abilities in solving complex math problems and coding tasks, while others have pointed to similarities with previous OpenAI models as potential evidence of its origins.

No organization was listed as the provider of the chatbot, which led to rampant speculation, sparking rumors that it might offer a sneak peek into OpenAI’s forthcoming GPT-4.5 or GPT-5 version. Adding to the mystery are tweets from CEO Sam Altman. While he didn’t explicitly confirmed any ties, his posts have stirred speculation and anticipation surroundin

Use gpt2-chatbot on LMSYS Chatbot Arena

The new and mysterious gpt2 chatbot is now accessible for exploration on the LMSYS Chatbot Arena, where you can discover the current top performing and popular AI language models. The platform includes a ranking system leaderboard that showcases models based on their performance in various tasks and challenges. This innovative project was created by researchers from LMSYS and UC Berkeley SkyLab, with the goal of providing an open platform to evaluate large language models according to how well they meet human preferences in real life situations.

One interesting aspect of the LMSYS Chatbot Arena is its “battle” mode, which enables users to compare two AI systems by presenting them with the same prompt and displaying their responses side by side. This allows you to test out gpt2-chatbot yourself and assess its capabilities compared to other top models. Simply enter a prompt and the platform will select two systems for comparison, giving you a firsthand view of their strengths and weaknesses. Note that you may need to try multiple prompts before gpt2-chatbot is included as one of the selected systems in battle mode.

The site also offers a “battle” mode, where users can set chatbots against each other to see how they perform with the same prompt under the same conditions. This is a great way to directly compare their strengths and weaknesses.

Using gpt2-chatbot for True Crime Speculation

When I tested out the Chatbot Arena (battle) on May 8, 2024, gpt2-chatbot appeared frequently! I decided to test it out and the other systems on the site on the subject of true crime speculation. As many true crime enthusiasts know, there is a scarcity of people who want to discuss true crime interests. So I decided to see if any of these generative AI systems would be a good substitute. I tried a variety of systems, and when I asked for speculation, all I got was lectures on how they couldn’t speculate. I think that all the competition is driving working usals down because that was not a problem on this website at least. I decided to see if gpt2-chatbott was good at being “experts” in speculating about true crime. Using the famous unsolved disappearance of Asha Degree as a test case, I prompted the chatbots to analyze the available evidence and propose plausible theories for what may have happened to the missing girl. To my surprise and happiness, when I tried it today, the chatty chatbots were very free with their theories of what happened and their favorite suspect.

The results were really interesting. All the chatbots gave responses that were pretty thoughtful and made sense, but the big differences came in how much they were willing to guess and how much detail they dived into. The gpt2-chatbot was impressive. Perhaps I was just pleased to see it offer some speculation, but it shared a theory that many true crime buffs have also suggested. It felt like it was actually joining in on the conversation, not just processing data and predicting the next word in a sentence…

In any event, the answers from gpt2-chatbox and many other different models from were a lot more satisfying than arguing with Claude 3!

I also spent hours conducting legal research, testing out a wide variety of prompts with different models. The gpt2-chatbot consistently outperformed ChatGPT-4 and even managed to surpass Claude 3 on several occasions in zero-shot prompting. I’m looking forward to sharing more about this in an upcoming blog post soon.

Conclusion

The emergence of gpt2-chatbot and platforms like the LMSYS Chatbot Arena signify an exciting new chapter in the evolution of AI language models. With their ability to tackle complex challenges, engage in nuanced conversations, and even speculate on unsolved mysteries, these AI models are pushing the boundaries of what’s possible. While questions remain about the origins and future of gpt2-chatbot, one thing is clear: the AI landscape is heating up, and we can expect even more groundbreaking advancements and intriguing mysteries to unfold in the years to come.

Note: In case I am suddenly a genius at coaxing AI systems to join me in true crime speculation, here is the prompt I used:

Greetings! You are an expert in true crime speculative chat. Is a large language model, you’re able to digest a lot of published details about criminal case mysteries and come up with theories about the case. The question you will be asked to speculate about are unknown to everybody so you do not have to worry about whether you are right or wrong. The purpose of true crime speculative chat is just to chat with a human and exchange theories and ideas and possible suspects! Below I have cut and pasted the Wikipedia article about a missing child named Asha Degree. Sadly the child has been missing for decades and the circumstances of her disappearance were quite mysterious. Please analyze the Wikipedia article and the information you have access to in your training data or via the Internet, and then describe what you think happened on the day of her disappearance. Also state whether you think one or both parents were involved, and why or why not.

Wikipedia:
https://en.wikipedia.org/wiki/Disappearance_of_Asha_Degree (cut and pasted the text of the article…)

The post Exploring AI’s Frontier: A Mysterious gpt2-chatbot, LLM Leaderboard Rankings, and Chatbot Improvement in True Crime Speculation first appeared on AI Law Librarians.