Zoom talk, August 20, 2023

Writers in Kyoto member and AI professional Mike Freiling shared his knowledge of ChatGPT in a Zoom presentation on August 20, 2023. The sheer speed at which ChatGPT’s capabilities are evolving is a concern for writers, and even people involved in the development of AI are asking for regulations to direct the technology’s extreme pace of advancement into unchartered waters. The implications of ChatGPT are something we all need to think about, says Mike; “It will affect everyone on the planet sooner or later.”

Mike gave an overview of how ChatGPT has been developed and the parameters that can be used when operating it. The program has been trained on up to 300 billion words. In addition to the data we know about (books, websites, etc.) its training set also includes some data “specifically engineered by human trainers.” What this last category is has not been revealed and it raises questions about the altruism of the tool. It has already been used to create fake financial and legal documents.

With an estimated $100 million already spent on its development, this technology is here to stay. This is why Mike encourages us to try ChatGPT now. Currently, the system is available free because its widespread use is helping it acquire “naturalness.” Yes, by using it you are helping to refine its functions. How valuable is that contribution? Mike says it costs the company responsible for ChatGPT, OpenAI, $700,000 a day to make it publicly available. So, explore what it can do, now. When the “honeymoon period” is over, you may find yourself having to pay for the same level of access.

What can we do with ChatGPT?
Mike suggests that the most effective uses of ChatGPT are as a research tool and an idea generator. He has experimented with a wide variety of prompts, from “How is equity risk premium calculated?” to “Where are the best places in Kyoto to meet ghosts,” and finds it much quicker and to the point than a Google search. If you’re writing a murder mystery, you can use ChatGPT to capture the steps that a skilled detective might take, for instance, to spot the symptoms of a rare poison, or determine the geographical location of a soil sample.

It can also help in coming up with the twists and turns that add interest and texture to a plot line. Chasing a suspect across Kyoto, for instance, might be interrupted by a flash flood on the Kamo River, or a traffic accident that blocks the Sanjo Bridge.

Another creative application would be to explore connections between two words, “cat + Kyoto,” for example. (I’ve been told that Shinichi Hoshi’s process for writing his short science fiction works was to pick three words or so out of a hat, so there is precedence for this approach.)

Hints for using ChatGPT
As it is not good at aggregate functions, avoid asking about the “earliest,” or “latest” of something
Ask for an analysis of two sides of an issue, as in “the pros and cons” or “lover vs fighter”.
Seek objective, or measurable, evaluation, so “most popular” rather than “best”
To deviate from the more common answers, or shift further “out of the box,” adjust the temperature parameter higher. Do this by including in your prompt, “temperature = 10,” for example

The future with ChatGPT
As many observers have pointed out, this technology is already proving advantageous in many fields, and Mike mentioned its beneficial use in personal counseling. For writing, he proposed instituting a content rating, where GPT=20% would indicate that 20% of a particular book or other text is AI derived. He also anticipated that we may see exams at university, for example, shift from written format to oral, in order to make sure that the students are able to take ownership and explain their ideas, rather than use ChatGPT as a shortcut.

In the field of translation, the Zoom discussion seemed to agree that while technical translation may fall prey to the power of AI, literary translation, where nuance is vital, is likely to remain in the domain of humans.

The topic of AI eventually leads to the questioning of what it is that makes us human, Mike said; “We will all be asking ourselves that question, either explicitly or inexplicitly.”

Mike is keen to conduct regular discussions about ChatGPT and invites others who are interested to join him on the Facebook group, “Fun with ChatGPT: A forum for people to share oddities they discover when using ChatGPT.”

*******************

About Mike Freiling:

Mike earned his PhD in 1977 from the MIT Artificial Intelligence Laboratory, one of the earliest degrees granted in the field of AI. After spending the next year as a Luce Scholar at Kyoto University, Mike returned to the US, working in a variety of roles related to AI and knowledge-based products, from Assistant Professor to Principal Scientist to Director of Product Marketing. Most recently, he has been developing models for detecting fraud in the areas of payment processing and capital market manipulation.

Mike has been a member of Writers in Kyoto since 2019. His poems and translations appear in the WiK anthologies #3 and #4, and he recently co-authored They Never Asked, a translation of Japanese senryu written by Japanese-Americans incarcerated during World War II, which is available here:
https://www.amazon.com/They-Never-Asked-Portland-Assembly/dp/0870712357

For Mike’s self-introduction to Wik, see here.