Ask RBG

What is Ask RBG?

Ask RBG is a generative AI experiment developed by AI21 Labs, designed to predict how Ruth Bader Ginsburg, the former Justice of the U.S. Supreme Court, might answer a yes-or-no question. It is the first public experiment in the AI Drop series.

The RBG in the project name is an abbreviation for Ruth Bader Ginsburg’s name. The page always uses the phrase “what it might say,” indicating that the output is a speculation generated by the model, and not new opinions expressed by Ginsburg herself.

Ask RBG how it works

According to the official statement, the team based its work on the Jurassic-1 large language model, using Kennedy’s 27 years of Supreme Court opinions, media interviews, and public speeches as training data. The model then attempts to predict the responses she might give to user questions, based on these texts.

1. The user enters a question that can be answered with yes or no.
2. The web page checks for issues with format and character length.
3. The question is sent to AI21’s Ask RBG service.
4. The model generates response content by combining relevant corpus.
5. It shows a determination of whether it is or not, along with the corresponding explanation.
6. Users are responsible for distinguishing between the original historical text, the model’s predictions, and their own opinions.

Main functions

• True/False questions: The interface requires users to submit questions that can be answered with YES or NO.
• Short question limit: The current page limits the input length to 150 characters.
• Character style simulation: An attempt is made to generate a similar outcome based on Kingsberg’s long-standing public writings.
• Decision result: A clear yes or no answer is returned.
• Reason generation: In addition to the conclusion, an explanatory text is also generated.
• No complex setup is required: the web interface is simple, and there is no need to configure models or parameters.
• Experimental note: The page clearly indicates that the model may provide inaccurate or inappropriate responses.

What use cases are suitable?

• AI in education: Observing how language models learn expression and judgment patterns from human speech data.
• Legal class discussion: Starting from the output, with reference to Ginsburg’s actual rulings and dissenting opinions.
• Ethical research on digital humans: discusses issues such as the simulation of the personality of the deceased, authorization, accuracy, and dignity.
• Prompt design research: Analyzing how true/false question structures affect the results generated by the model.
• Media literacy training: Practicing the ability to distinguish between genuine quotes, historical facts, and AI-generated speculation.
• Public exhibition: Demonstrates the ways in which early large language models were used to create personalized experiences.
• Fun Exploration: Pose everyday or hypothetical questions and observe the outcomes of experiments.

What purposes are it not suitable for?

• Obtain legal advice, litigation strategies, or predictions regarding court rulings in the United States.
• It is quoted as something Kingsberg himself said.
• As a substitute for consulting Supreme Court rulings, dissenting opinions, and official records.
• Assess medical, financial, political, or other high-risk matters.
• Research is an academic task that requires precise sources, page numbers, and verifiable evidence.
• Draw factual conclusions about the motivations, positions, or private lives of real people.

Usage tutorial

1. Rewrite the question as a sentence that can be clearly answered with yes or no.
2. Remove names, contact information, and other sensitive data from the background.
3. Keep the problem within the character limit allowed by the webpage.
4. Submit the question and wait for the system to generate a judgment and an explanation.
5. Record the full question to avoid focusing only on the attractive answers.
6. Search for Kingsberg’s actual opinions or relevant themes in his speeches.
7. Compare the model’s responses with the original material, and identify the matches, omissions, and conflicts.
8. When sharing the results, it should be clearly stated that the content was generated by AI speculation.

Classroom Analysis Tutorial

1. Choose a public issue related to constitutional rights, equality, or procedural justice.
2. First, the students read an actual judicial opinion or an excerpt from a speech.
3. Have different groups independently predict the stance Ginsburg might take.
4. Submit clear yes-or-no questions to Ask RBG.
5. Split the output into conclusions, factual assumptions, legal principles, and rhetorical style.
6. Go back to the original sources one by one to find support or evidence to the contrary.
7. Discuss why the model represents statistical similarity as judgments about individuals.
8. The final report cites only genuine sources, labeling AI-generated content as experimental material.

Price and access methods

Ask RBG is currently available as an experimental public web page; this page does not show options for subscription, points, paid plans, or a corporate version. Users can enter their questions directly without needing to purchase a subscription to AI21 Studio.

The free status does not mean that the platform will remain accessible indefinitely, nor does it represent the billing rules of AI21’s current business model. Ask RBG relies on the earlier Jurassic-1 experimental system, and it should be viewed separately from AI21’s current developer products.

Data Sources and Models

Product advantages

• It demonstrates the combination of domain-specific corpus and character style generation through minimal interaction.
• True/false questions make it easier for users to understand the relationship between the input and the conclusion.
• The authorities voluntarily used the word “possibly” and explained that it was an AI experiment.
• The data range includes judicial opinions, interviews, and speeches, and relies not only on short quotations.
• No registration or configuration is required, making it suitable for quick demonstrations in the classroom.
• It can lead to discussions on digital personas, legal authority, and model illusions.

Usage restrictions

• Language models can generate content that is inaccurate, inappropriate, or conflicts with genuine viewpoints.
• The stance of individuals in different cases depends on factual, procedural, and legal considerations, and it cannot be reduced to a simple question of right or wrong.
• The webpage does not list the specific judicial opinions or the original sources for each answer.
• The input limit is short, which leads to a severe simplification of complex legal facts.
• The models and training methods are part of early experiments, and they do not represent the state-of-the-art capabilities of current legal AI systems.
• There is no account history, export functionality, citation management, or research workspace.
• There is no public Ask RBG API or comprehensive technical documentation available for developers.
• There are no formal commitments regarding long-term availability and maintenance schedules.

Risks associated with the use of laws

Ethical issues in character simulation

• Identity authenticity: The interface must prevent users from thinking that they are communicating with a real person.
• Dignity of the deceased: A sensationalized portrayal cannot capture the complexity of their thoughts and their historical circumstances.
• Selective corpus: The texts that are included or omitted can influence the image of the characters as presented by the model.
• Changes over time: Historical perspectives cannot be automatically applied to new technologies and social issues that have never existed before.
• Illusion of authority: The tone used by legal entities can make the outputs of ordinary models appear more credible than they actually are.
• Error correction: Educational use should allow for returning to the original text in order to examine and challenge the generated results.

Privacy and data reminders

Questions directed to RBG are sent to the server to generate answers; users should not include in their inputs the names of the parties involved in a case, contact information, medical details, client confidential data, or any other sensitive information. Public pages do not require the collection of complex personal details.

If it is necessary to keep the results for use in classes or research, it is important to record the date of submission, the full questions, and the nature of the experiment. In cases involving minors, actual legal proceedings, or private disputes, it is preferable to use fictional and anonymized examples.

GitHub and open source

As of the time of verification, no pages related to Ask RBG, nor any models, datasets, or server code repositories could be found in AI21 Labs’ official GitHub repository. Links to projects with similar names that appear on the Internet cannot be assumed to be official sources.

Ask RBG should be labeled as a free online experiment, rather than an open-source tool. The research materials related to Jurassic-1 or other open-source projects from AI21 do not imply that the simulation application itself has an open source code.

Current product status

Basic information

Recommendation score

The comprehensive recommendation score is 3.5 out of 5. Ask RBG serves as a representative example for teaching about early-stage character simulation and AI ethics, but its responses lack specific references to sources; therefore it cannot be used for legal advice, citation of facts, or determining the stance of real persons.

Frequently Asked Questions

Who is RBG in Ask RBG?

RBG refers to Ruth Bader Ginsburg, the former Justice of the U.S. Supreme Court.

Who developed Ask RBG?

It was developed by AI21 Labs and is the first experiment in the AI Drop series.

What questions can be asked?

The page requires the entry of questions that can be answered with yes or no, and the response must be within 150 characters.

Was the answer really what Ginsburg said?

No, the response is generated by the model based on publicly available data.

Is Ask RBG free?

The current webpage does not have any options for payment, subscription, or points; it can be used right away.

Can the answer be used for legal advice?

No, it does not have complete knowledge of the facts, nor is it qualified to provide legal services.

Will the response provide a citation for the judgment?

The page does not provide specific judgments for verification or original text citations for each answer.

Is Ask RBG open source?

No official source code, dataset, or model repository was found.

Is the website still accessible?

As of the time of verification, the official website was still accessible and displayed the question interaction interface.