People are super excited about generative AI tools like ChatGPT. Here is some info about these tools at JHU. This is a fast-moving area, with features and availability changing almost weekly, so WSE affiliates should please contact us at [email protected] if you want the latest. If you’re not WSE affiliated you should contact your local IT support team (we still think you’re awesome, but we won’t be able to help you out).
BE EXTREMELY CAREFUL about what information you use with generative AI tools! Most of them will use the information you provide to them for additional training information, which may result in your content being replayed to future generative AI users. DO NOT USE PHI / PII / JH proprietary information with any of these tools unless you are absolutely certain of how your provided information might be saved or used.
- These tools can seem like magic1. Understanding a little about how they work can help you know what kinds of problems they might help with and where the answers come from. I suggest this quick introduction as a way to understand what you’re getting: https://www.oneusefulthing.org/p/thinking-like-an-ai. Here’s another article that gives 15 things that GPTs are good at and five where they are not: https://www.oneusefulthing.org/p/15-times-to-use-ai-and-5-not-to.
- Which of these tools is the “best?” It’s very subjective and depends on what you are trying to accomplish. If you want to lean into the subjectivity, the Chatbot Arena records people’s impressions of how different tools answer the same questions.
- Hopkins’ IT@JH is developing a tool that will allow for the use of multiple AI models (ChatGPT, Claude, etc) inside a safe environment. The system aims to provide models that can be grounded with information provided by the users and that can be shared between teams. As of October 2024 the prototype is being demonstrated, and it is due to enter user testing in early 2025.
- There is not an existing agreement with OpenAI directly for ChatGPT. We know many people have personal licenses for ChatGPT, but those are personal arrangements even if bought with JH funds. Do not use JH proprietary information or PHI / PII with tools paid for to OpenAI. You are personally liable for any consequences from entering into a personal agreement with OpenAI.
- JH has opened access to secure, user-friendly options like Copilot for Microsoft 365 and Teams Premium. There is be a recurring per-user flat fee to access these services, which are available in the MyJH software store. (Here is an article comparing the two products.) Any type of information (PHI / PII / JH proprietary) may be used in Copilot for Microsoft 365.
- JH also has an agreement with Microsoft’s Azure cloud compute platform, and OpenAI tools are available inside Azure. This Azure service is useful for technically inclined users like data scientists or programmers, though the Azure AI Studio environment is reasonably user friendly (see screenshot below). The OpenAI tools need to be enabled in your Azure instance, so if you already have a JH Azure instance you will need to request them to be added. Azure use is consumption based, so you only pay for the amount of AI processing you use, not a monthly fee. Any type of information (PHI / PII / JH proprietary) may be used in OpenAI tools in Azure.
- JH has an agreement with Amazon’s AWS cloud compute platform, and in their Bedrock service you can get access to models like Claude and Llama.This AWS service is targeted at programmers and won’t be useful to most people without some sort of front-end client or custom programming. AWS use is consumption based, so you only pay for the amount of AI processing you use, not a monthly fee.
- Zoom’s AI Companion is available to all WSE Zoom users.
- Other generative AI tools are available in Adobe Creative Cloud, for example Adobe Firefly. Creative Cloud is available as a subscription in the software section of MyJH.
- IT@JH has a page about AI at JH.
- Here’s a really interesting prompt you might want to try in one of these tools… It’ll provide you a summary of exactly what you want to hear your superpower is. The basic underlying tuning on these engines is that they’re supposed to be supportive collaborators, so instead of insight you’ll get a reflection of the information you provide it.
“Each human you speak to has a single, specific superpower, and it is your job to help that person identify it. These superpowers are more subtle than those portrayed in comic books-think of them as “what each person is better at than anyone else in the world.” These may not be obvious things, as there is only one person who is “the best golfer in the world” or “the fastest person in the world.” Most people’s superpowers lie in nuanced, unique intersections of skills, interests, and experiences. You will interview me by asking up to 15 questions about what I enjoy, excel at, or find meaningful. Then, you will suggest five possible areas where my superpower might lie. At this stage, you are not defining my exact superpower but exploring its nature or domain. Once I choose one of these areas, you will guide me to further analyze what exactly my superpower might be. This process should feel organic-ask one question at a time, and avoid explaining the whole exercise or sending all questions at once. Please remember: do not exceed 15 questions, and after I answer the 15th question you ask me, transition automatically to the next step of identifying the potential areas where my superpower might reside.” ↩︎