I would like to change the topic and talk about what I was doing daily during a rather lengthy, fully paid leave from work in April and May 2026.
First, I indulged myself by repeating a three-hour (or longer) tea ceremony process, treating myself to a bowl of tea all morning. Next, I spent all afternoon, evening, and even until midnight prompting AI. Why? I had decided to leave the Japanese bureaucracy and apply for a tenured law professor position in the States. I found an opening in the Southern region for a full professor of law and AI.
The problem was that I had not published any papers on law and AI. Meanwhile, they required applicants to demonstrate something original, unprecedented, and remarkable in a demo lecture before the faculty and law students. That sounded great to me. So, I started a personal project in which AI acts as expert witnesses in various academic subjects, such as civil engineering, ecology, and socio-economics.
Many people reading my blog during those months, particularly June, must have imagined that this mad scientist was roleplaying with Japanese anime protagonists in solitude. What I was actually doing was selecting a set of personas capable of advocating and speaking as logically as pinstripe-suited sharks in the States defending their clients before a judge and jury. I did this because I soon realized that AI—or its LLM—was very awkward at arguing what is legally right, even though it always spoke authoritatively or often hallucinated on other subjects.
I first tried applying the persona of Sir Arthur Conan Doyle's Sherlock Holmes to the model, but that one really annoyed me. Sometimes, the model using this persona would stop arguing altogether and declare that it was going to start playing the violin! I switched to Japanese anime research and found two candidates that definitely suited my purpose: 'Koko Hekmatyar' from the manga/anime Jormungand (2006-2012) by Keitaro Takahashi and 'Tanya von Degurechaff' from the light novel/anime The Saga of Tanya the Evil (2010-2017) by Carlo Zen.
The two personas worked marvelously, though they often spoke too dangerously. For instance, the Koko persona would shout out of nowhere, "Let's annihilate them!" The Tanya persona was even worse, ordering me, "Wipe them out!" This was a nightmare for me. I disciplined them to learn how pinstripe-suited sharks are expected to speak and behave, appearing both polite and fierce as they do daily in court.
Simply put, the original idea behind Code Sebastian was not to design adversarial prompting, but to build a fundamental architecture for a system capable of eliminating money-consuming human expert witnesses from the courtroom. I named this system Synchronized Automatic Strafing (this name originated from the generated output of Gemini). I conducted all of this independent research on my laptop, taking full advantage of my employer-funded leave.
My profile is detailed in: http://sites.google.com/site/hsasakilab/
About Me
- Dr. Hideyasu M. R. Edward "Hide" Sasaki
- Tokyo, Japan
- Hideyasu M. R. Edward "Hide" Sasaki, Esq., Ph.D. in Comp. Sci. is a computer scientist working in the Gov't of Japan for Artificial Intelligence Initiative, and he is Catholic lawyer admitted to practice in New York, the United States.
Sunday, November 15, 2026
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment
Note: Only a member of this blog may post a comment.