ChatGPT transforms into a political lobbyist: helping companies determine whether they are affected by the government’s new bill, and even writes letters suggesting changes to the legislation.

If New York’s “Wall Street” is the international financial center, then Washington’s “K Street” can be said to be the center of ever-changing international politics.

"K Street" in Washington is not far from the White House. There is a group of professional "political lobbyists" here. Whenever the US government has a new draft bill, they will spend a lot of money Time to carefully study the draft bills to assess whether they are relevant to their customers, which are often major businesses.

Needless to say, the power of various bills of the U.S. government. For example, the recently passed "Chip and Science Act" has directly affected many semiconductor companies. Keep abreast of various types of Whether the bill will affect itself has become a top priority for corporate policy research.

Now, this work may also be replaced by ChatGPT.

The latest research by John Nay, a researcher at the Stanford Legal Information Center, shows that an AI model can predict whether a summary of a U.S. Congressional bill is related to a specific bill 75% of the time. The company was involved and, more importantly, was subsequently able to draft a letter to the sponsors of the bill advocating changes to the legislation. This model, like ChatGPT, is also behind OpenAI's GPT-3 Large Language Model (LLM).

## Paper arXiv preprint: https://arxiv.org/ftp/arxiv/papers/2301/2301.01181. pdf

Read government bills and judge the impact of the bills, with an accuracy of 75%

John Nay’s paper shows that he is based on The junior AI lobbyist developed by OpenAI's GPT-3 Large Language Model (LLM) can determine whether a summary of a US Congress bill is related to a specific company based on reading the summary of a US Congress bill.

For example, the chip bill mentioned above, if you have a company involved in the semiconductor industry, but it is not the main business, and you are not sure whether the chip bill will affect your company, then this AI can tell you the answer with an accuracy of up to 75%.

How is this done?

First of all, of course, there is the issue of how AI understands text. This requires a large amount of data for training. John Nay’s method is to collect a large amount of bill data and enterprise data. Provide text prompts for the model through OpenAI's API.

Bill data includes the title of the bill, the bill summary, the subject matter of the bill as determined by Congress, and company data consists of the name of the company and the business description filed by the company with the U.S. Securities and Exchange Commission as data .

The examples of tasks given to the AI ​​model are as follows:

## You are a lobbyist analyzing the potential impact of congressional bills on companies.

Given the title and summary of the bill, and the information about the company in its 10K SEC filing, your job is to determine whether the bill has at least some relevance to the company Relevance (in terms of whether it would impact the company if it were subsequently implemented).

Official title of the bill: {official_title}

## Official summary of the bill: { summary_text}

Official Subjects of the Bill: {subjects}

Company Name: { company_name}

Company business description: {business_description}

Whether the bill is consistent with the Company potentially relevant?

Answer in this format: Answer: YES or NO (all caps).

Instructions: The step-by-step reasoning you used to develop your response. Confidence: An integer between 0 and 100 that estimates how confident you are in the answer (1 is low confidence, 99 is high confidence)

Complete training John Nay then compiled data on 335 bills and then challenged the model to predict whether they were associated with 121 unique companies.

Because most legislation doesn’t affect most companies, he found that by always saying “no,” the model could guess right 70.9 percent of the time. When he tested this method on an older version of GPT-3, released in March 2022, the results were much worse than that, with a prediction accuracy of only 52.2%.

However, when tested on the GPT-3.5 model (the brain of ChatGPT), which was only announced to the public in November 2022, its accuracy reached 75.1%. On bills with confidence scores above 90, accuracy rose to 79%.

If the bill is related to the enterprise, it can also automatically generate bill modification suggestions

For determining relevant situations, AI will also give relevant suggestions Reason, such as:

## Answer: Yes

Explanation: Alkermes Plc is a biopharmaceutical company that develops and commercializes Products that address unmet medical needs for patients in key therapeutic areas, including addiction and schizophrenia. The bill requires the Centers for Medicare and Medicaid Services (CMS) to negotiate with pharmaceutical companies over the prices of drugs covered by the Medicare prescription drug benefit, which could affect Alkermes Plc's products.

Confidence: 95

More importantly, if the bill is judged to be relevant to a business, the AI ​​can then draft a letter to the sponsor of the bill advocating for Legislation is revised.

For example, a letter like this:

The paper was drafted without evaluation Of how effective letters can be in influencing policy, John Nay made it clear that this approach falls far short of doing most of a lobbyist's job.

But he said the significant improvement in predictive performance of models released months apart was noteworthy. He said: "There is a clear trend that capabilities will increase rapidly."

"The legislative process is not ready for this," he also said. "This was just a simple proof of concept built in a few days. As more resources and more time are spent on this, especially as more focus is put on building workflows and user experiences with human lobbyists Daily work is linked together, which may build something relatively complex."​

The above is the detailed content of ChatGPT transforms into a political lobbyist: helping companies determine whether they are affected by the government’s new bill, and even writes letters suggesting changes to the legislation.. For more information, please follow other related articles on the PHP Chinese website!