Artificial Intelligence (AI) refers to the capability of machines to perform tasks that typically require human intelligence. AI systems learn from data, make decisions using algorithms, and interact with humans through technologies like speech recognition.

Generative AI (GenAI) is a branch of AI that uses patterns in data to create digital things like text, images, and videos. These tools can provide answers but also iterate upon ideas over the course of a conversation. Unlike AI tools before them, GenAI tools such as ChatGPT have been adopted on a massive scale, including in educational settings.

AI models like ChatGPT are called large language models, or LLMs. They have been trained on vast amounts of data from the internet from sites like Wikipedia and Reddit. By processing the patterns in this data, they have become very good at predicting the next word or words in a sentence. In a sense, you can think of LLMs like word completion on your smartphone, but on a larger scale where it can compose entire essays. However, this is a simplified overview of how LLMs work, so read below to better understand the nuances of how LLMs operate and where they fit for you.

• Do try it out yourself. Everyone’s experience will be unique because LLMs produce slightly different results for each user. Try out some AI models for yourself and see how they match or don’t match your expectations. Stanford’s approved AI Playground is a great place to test out the exciting capabilities of AI in a semi-secure environment.
• Do consider how to use AI tools responsibly. You should never input high-risk data, such as credit card numbers or sensitive research data, in AI environments unless they have been approved for that level of security. You should also be cognizant of inputting medium- or low-risk data, and be mindful of whether this information is approved to share broadly.
• Do familiarize yourself with Stanford’s guidance for AI use in the classroom. Per the Office of Community Standards, “absent a clear statement from a course instructor, use of or consultation with generative AI shall be treated analogously to assistance from another person.” In addition, it is generally a good practice to be transparent and open about the role that AI services or tools play in your work or interaction with colleagues or learners.
• Don’t rely on AI’s results and sources to be accurate and factual. Many AI tools operate on probability, so the output you get may not be 100% accurate; instead, it may just be the best available response, which can still be wrong. It is a good practice to check the output of AI tools before moving forward.
• Don’t include any personal or confidential information when using AI. As with any non-secure technology, it is a good practice to not put any information into the tool that you would not be comfortable having on the public internet. Your privacy is not guaranteed, and in some cases the information may be used by the AI provider to refine its technology.

AI systems collect data from various sources either online or from information you share (i.e., prompts), including numbers, text, images, and sounds. It will analyze this data to learn patterns and relationships, helping the AI to make decisions or predictions. Over time, AI improves its predictions by processing more data, which in turn will help it be more accurate and efficient.

Usage data is the information gathered when you use the tool. This can include your conversations and account information linked to the tool. The AI provider may use this data to train the model further. All this data may be stored in the system, which can be safe but be aware of data breaches, selling of data, and when you remove your account. You can generally find out how the data is stored, what data is stored, and if it is used for training in the terms of service agreement.

University IT (UIT) has set up an AI Playground. This is a terrific sandbox to try out various chatbots and get a feel for how each performs. You can attach documents in this environment, too. UIT notes that AI providers have committed to not retaining user data in the AI Playground. However, they encourage inputting only low-risk data.

Also, UIT’s GenAI Tool Evaluation Matrix site lists AI tools currently being evaluated for potential implementation in various contexts at Stanford; however, UIT advises that “[t]his page listing does not imply that each tool or program is approved for general use at Stanford.” For more information, you can refer to UIT’s Responsible AI at Stanford page which includes a comprehensive overview of considerations when using AI.

The cost of AI varies widely. Many consumer-facing AI tools, such as ChatGPT, may be free to use with limited AI models. Often, these providers also offer paid access to more advanced AI models with expanded capabilities. You may also discover AI tools that are integrated into products such as productivity suites or research engines; these often start out free-of-cost, but are limited in usage until you pay for a subscription or license. Finally, for individuals with more technical skills, it is also possible to build custom AI tools with APIs that connect to services like OpenAI. Ongoing costs of those custom builds can include maintenance, updates, and scaling the system to handle more data or users.

AI can at times be a “blackbox” in terms of how it works and what data it was trained on, so it’s difficult to pinpoint what data might have been included. You might consider using services that prioritize transparency in how they collect data and how they train their AI models.

In some situations, AI can offer a shortcut to points of friction in the learning process. In these cases, it is important to be mindful on how you want to use AI and to be clear with others on your standards. A clear syllabus statement can be a very effective tool in addressing this. In addition, if you are in conversations with students, emphasize the reasoning behind the instances in which you discourage or encourage the use of AI. For more, see this resource from the Stanford Center on Teaching and Learning.

AI is known for making errors, sometimes called “hallucinations,” such as incorrect facts, bias, and nonexistent citations. We recommend that you are mindful to proofread and fact-check anything that you produce with AI.

Some AI tools work differently on different platforms, including between Mac and PC, as well as mobile and desktop. We recommend testing out your tool on any device you intend to use it on, including mobile if you want others to have access to it.

University IT and Stanford’s Information Security Office (ISO) provide several resources on technology policies and digital tool access. You can learn more about the tools they are currently evaluating and how they recommend approaching AI at Gen AI Tool Evaluation Matrix and Responsible AI at Stanford, and about how to effectively use AI via their articles on GenAI Use-Cases for Experimenting and GenAI Prompt Guide.

Also, if you are considering purchasing software or hardware (whether or not it is AI-related), we recommend reading UIT’s guidance on this topic. For academic-related policies, the Office of Community Standards also provides a resource about Generative AI Policy Guidance and how it affects the honor code.

If you have specific questions pertaining to your role at GSE, please see the personas in the section below.