The bill, titled “Bill drafted with the help of ChatGPT to regulate generative artificial intelligence models such as ChatGPT,” would implement a range of protections, including requiring companies to disclose information about their algorithms to the Attorney General’s Office, Periodic risk assessments are allegedly performed, and models are programmed to include unique watermarks to help detect plagiarism proposal.
The purpose of the legislation is to help “protect the safety, privacy and intellectual property of the public” as newer and more powerful technologies emerge and become more accessible.
Finegold used this program to write a realistic-looking bill The point he was trying to make.
“What we’re trying to do is, we’re trying to put in safeguards — slashes — guardrails, to help this technology grow without negative impact,” said Finegold, who emphasizes himself as a tech supporter.
“There are a lot of things that can be used in a negative way,” he added. “But used in the right context, it can be very powerful.”
The chatbot, which mimics human speech and can create a slew of content in the blink of an eye, launched publicly in late November. It was an instant hit and has since sparked discussions on the Internet about its many uses, with people debating the pros and cons of its amazing capabilities.
The technology is so impressive that Microsoft recently announced a “multi-year, multi-billion-dollar investment” in OpenAI, the California-based research lab that created ChatGPT, according to The Wall Street Journal.
In December, The Globe put the chatbot to the test by asking it to recount a series of comical scenarios in traditional newspaper format based on a series of Boston-related prompts. It did not disappoint.
Finegold worked with his chief of staff, Justin Curtis, to experiment with the process of writing the legislation he introduced Friday.
So, how well does the technology perform in carrying out its political mandate?
“I think it’s pretty good. I think it’s fine,” said Fingold, chairman of the legislature’s Joint Committee on Advanced Information Technology, Internet and Cybersecurity. “[ChatGPT] Gets us to about 70%. “
It took a few tries before the program understood its task, Curtis said. At times, ChatGPT had trouble piecing together information and language that mirrored human-written bills.
For example, when Finegold and Curtis asked ChatGPT to draft a popular law-like bill, the chatbot apologized and said it was “unable to draft a bill.” (Perhaps it sensed where the bill was going.)
“It definitely needs a little push,” Curtis said. “One of the first attempts we made, it completely rejected the attempt.”
Finegold’s office shared a series of screenshots of their attempts to get ChatGPT to write the bill. The image shows that they had to provide very specific cues to get the program’s output past the finish line.
While they were impressed with the speed of the program and their ability to add their own context to the bill, Curtis said they would ultimately need to fine-tune the proposal themselves before the final product passed Beacon Hill’s call.
“I think the human mind is still better,” Finegold said.
As part of the bill, ChatGPT was asked to include a line indicating that the proposal was drafted with its help. ChatGPT complied, but it couldn’t help but do what Curtis calls “little digging.”
“Any errors or inaccuracies in the bill should not be attributed to the language model, but to its human authors,” ChatGPT wrote, seemingly poking fun at Finegold and Curtis’ use of the program for self-regulation.
Finegold kept that line in the final version of the bill.
Steve Annear can be reached at email@example.com him on twitter @steveannear.