FAQ

HopGPT is a generative AI platform that provides members of the Johns Hopkins University and Medicine community with secure, JHED-authenticated access to multiple Large Language Models (LLMs) like ChatGPT and Claude through a familiar interface.

Your chat prompts and AI-generated responses are purged after 30 days. Please review our Data Retention Notice for more information.

The Sheridan Libraries has assembled a number of AI literacy resources. Please visit Artificial Intelligence for Research & Scholarship.

Models options available through HopGPT can (and do) change frequently. Each has its own strengths and weaknesses. When making your selection, we include subtitles to highlight each model’s strengths. A web search, given your use case, for example, “best llm for document summary,” might reveal GPT-5 or Claude Sonnet 4 as an excellent choice. Please visit Choosing a Model in HopGPT for more information.

While HopGPT is approved for sensitive data, including PHI and PII, you must act in a manner consistent with appropriate use of PHI / PII and be a responsible steward of the data you enter into and collect from HopGPT. In accordance with Johns Hopkins University and Johns Hopkins Medicine policies, you should always minimize the amount of sensitive information used to accomplish your intended goal.

HopGPT is safe to use with sensitive data, including student information, but should not be used to perform student-related academic assessments as inputs and outputs are only retained for 30 days. Further, you are responsible for acting in a manner consistent with the Terms of Use and applicable Johns Hopkins University and Medicine policies.

The exact file size, or context limit, varies depending on the specific implementation. Generally, files up to several MB can be processed, but larger files may need to be split or summarized.

LLM Context Limits:

• Context limits vary by model, typically ranging from about 2,000 to 32,000 tokens.
• A token is roughly 4 characters or 3/4 of a word in English.
• For example, GPT-3.5 has a limit of about 4,000 tokens, while GPT-4 can handle up to 8,000 or 32,000 tokens, depending on the version.

Longer documents may need to be summarized or split into smaller chunks. HopGPT can analyze large files but may not be able to hold the entire content in its “memory” at once.

ChatGPT Edu is available to faculty and staff in the Whiting School of Engineering. Key differences from HopGPT include:

• Access to additional OpenAI models
• Ability to create and share custom GPTs or assistants with your team or department

ChatGPT Edu requires an annual subscription and does not provide access to models outside of the OpenAI portfolio. For more information about this service, please contact [email protected].

HopGPT is a product governed by several organizations. To request consideration of an additional model, email [email protected]. Be sure to include your use case in your correspondence.

API access to models available within HopGPT is considered on a case-by-case basis and does require a cost object. If you would like to request access, please contact us.

The product team is working to develop repeatable processes to support custom assistants. We expect this to be the focus of our efforts into 2026. Please share your use case with us to help prioritize our efforts.

What is a context window limit?

A context window is the amount of conversation an AI model can actively “remember” at one time. When a discussion becomes very long, earlier parts may fall outside this window and are no longer considered when generating responses. 

Why does this happen?

This is a technical limitation of the AI model itself. Every AI model has a maximum amount of text it can process at once. Once that limit is reached, older content must be dropped to make room for new messages. 

Is this a HopGPT limitation?

This behavior comes from the underlying AI model, not HopGPT. HopGPT simply provides an interface for interacting with the model. 

What might I notice when the limit is reached?

You may see things like: 

• The AI forgetting earlier details 
• The AI asking for information already discussed 
• Responses that seem to ignore older context 

This does not mean anything is broken — it simply means earlier messages are outside the active window. 

How can I work around this?

HopGPT provides Forking tools that allow you to restart the conversation while preserving relevant context.

Why does forking help?

Forking effectively “resets” the context window. It gives the AI a fresh working memory while keeping the information you choose to preserve. 

Best Practice Tip

For very long or complex discussions, periodic forking can improve response quality and consistency.

Fork: Visual Messages Only

Use this when:

• Your conversation relies heavily on screenshots or diagrams 
• You want to remove long text history

Fork: Include Related Branches

Use this when:

• You have multiple connected threads 
• You want continuity without clutter

Fork: Include All to/from Here

Use this when:

• The conversation is very long 
• You want a clean restart with full context

Our platform supports a variety of commonly used file formats. Below is a quick guide to each file type, what it’s typically used for, and any important notes.