ChatGPT - bblockwood/lab GitHub Wiki

Wharton's ChatGPT Edu account

Logging into your account

Full-time staff member's can sign up for a ChatGPT Edu account through Wharton, which includes everything that a free account offers plus several enhancements. These include, but are not limited to, enhanced data privacy, access to improved models, custom GPTs, API access, and more processing power and tokens.

To log in to your ChatGPT Edu account:

  1. Access the web version of ChatGPT here or download one of the ChatGPT apps here.
  2. Click the Log in.
  3. Enter your email address as pennkey@upenn.edu.
  4. Click Continue.
  5. You will be taken to the PennKey login screen.
  6. Enter your PennKey username and password.
  7. Confirm with Two-Step.
  8. You are now logged into Wharton's ChatGPT Edu workspace.

Data Security

The University of Pennsylvania ensures that no data given to Wharton's ChatGPT Edu workspace is used to train the AI models, permitting you to upload data to your workspace that falls within certain approved University data classifications, specifically those at or below a defined sensitivity level. This allows you to upload data that is categorized as having low or moderate sensitivity, in accordance with government regulations and University policies, as long as you are not using Personally Identifiable Information or FERPA data. If you are, you must engage with Wharton Information Security Office (ISO) for a risk review of the overall initiative’s data architecture and flow outside of the AI solution. If you are using data categorized as high sensitivity or HIPAA/Protected Health Information data, you may not upload it to your ChatGPT Edu workspace.

For more information on security and data, go here. For more information on Penn data risk classification, go here.

API access

If you are working on a project that requires API access and would like to use ChatGPT, please contact your Wharton Computing Representative, as API access is currently in the pilot stage. You can go here if you would like additional information on API security information.

For additional information on Wharton's ChatGPT Edu workspace, go here.

Best practices

Prompt engineering

A prompt is a text, image, audio, data, or document input that initiates a conversation with a Large Language Model (LLM). When using ChatGPT for coding or research tasks, a well-written prompt clearly defines the problem or question, includes relevant context, and outlines any specific goals or constraints. For example, if you're troubleshooting a piece of code, it's helpful to include the language, a snippet of the code, the error message, and a short explanation of what you're trying to achieve. The more targeted and informative your prompt, the more accurate and relevant the model's response will be.

Prompting is often an iterative process. You might begin with a rough idea of what you need, review the model's response, and then revise your prompt to sharpen the focus or add clarity. For research-related queries, this might mean refining the scope of a literature summary, rephrasing a complex question, or adding background information on the topic. Whether you're looking for clean code, an explanation of a concept, or help organizing technical content, thoughtful prompt construction is key to getting high-quality results.

For additional information on prompt engineering, go here.

ChatGPT "hallucinations"

ChatGPT is a generative model, not a factual database. Because of this, it will sometimes produce an output that sounds plausible but is incorrect. This is sometimes referred to as ChatGPT "hallucinating."

In coding, this often presents itself as a ChatGPT-generated code output containing commands, options, or syntax that do not work and produce errors. Sometimes, however, it will produce code that runs without errors but does not do what you intend the code to do. If the code does not work, rephrasing your question or providing more context may help you get an output that is helpful, but ChatGPT will oftentimes continue to produce an answer for a problem even if an answer does not exist. It is critical that you test the code and ensure the results are logical. If you take ChatGPT outputs as given, you may make mistakes in your code.

This is not only a concern for coding-related prompts. This is true for other general knowledge or research-related questions you may ask. It is important to double check information given to you by ChatGPT and to make sure that sources and citations that it gives you exist.

Deep Research

Deep Research is a specialized ChatGPT feature that performs multi-step, web-based research. Unlike a regular conversation with ChatGPT, Deep Research is designed to find and synthesize information from a wide range of public web sources to answer complex, open-ended questions. If you just need a quick answer—like a definition, statistic, or update—Deep Research is usually unnecessary.

This feature is especially useful for research tasks that would typically require searching through multiple websites, reading across sources, and organizing that information into a coherent summary. Deep Research is a good starter step in conducting a literature review, although it is not a replacement for one. You should check the sources Deep Research gives you as a way to find more sources. You should also check other databases or use Google Scholar to search for additional literature.

Once you enter your question and select “Run deep research” under the "Tools" tab, ChatGPT will sometimes ask follow-up questions or generate a short form to clarify what you’re looking for. Then, it searches the web, reads pages, and constructs a detailed report. This process usually takes 5 to 30 minutes, depending on the complexity of the task. You’ll see a sidebar during the task showing the sources being visited and the model’s thought process.

The output will include a written report, along with citations and links to sources so you can verify or follow up on any claims. You can also attach files (such as PDFs or spreadsheets) to give more context.

Note that deep research can only access publicly available information. It cannot access subscription-based content or private databases. If you’re working with paywalled or proprietary sources, you’ll need to provide that content manually.

For additional information on ChatGPT Deep Research, go here.