What PennChat Is

• Penn's own AI chat portal, hosted by the University
• Useful as an AI ‘baseline’: solid chatbot interface, Penn approved
• One interface, two model families: Claude and GPT
• Cleared for Low, Moderate, and most High Risk Penn data

Important: Needs PennNet or GlobalProtect.

Pros and Cons of PennChat

• Solid chatbot functionality, with relatively-current models
• “Agents” are custom GPTs, but yours must be rebuilt here
  • Important: you cannot share PennChat Agents (yet)
• No web search: it cannot look anything up or check a live source
• Chatbot ONLY, nothing like Claude Cowork or Code
• Noticeably slower than direct access, and results may differ

A good baseline. Not a replacement for direct access.

How It Compares

*Long-term PennChat pricing/financing is TBD

Demo: Let It Argue With You

A working draft, and a hostile referee

• Attach a draft you are working on
• Ask it to read as a hostile referee and to say what would fix each problem
• A long draft takes a couple of minutes to read
• The point is the fixes it names, not the praise it withholds

Which Model to Pick

• Start with Claude Sonnet 5, and pin it so it sits at the top of the picker
• It handles nearly all your work well
• The picker groups models into Premium, Balanced, Economical, and Legacy
• Model names change as new versions ship; the tiers do not
• Premium models draw more of your usage credits

Pick the newest model in the tier that fits the task.

Demo: The Model Picker

Model picker, expanded

• Open the picker at the top of the chat
• Premium, Balanced, Economical, Legacy
• Pin Claude Sonnet 5 so it is always the first thing you see

Agents :: Custom GPTs

• An agent is a chat you have set up in advance:
  • instructions it always follows
  • knowledge files it can search
  • tools, including a code interpreter
• No direct import: paste the instructions, re-upload the files

Most capabilities carry over, but web browsing does not.

Demo: Build a Virtual TA

Agent Builder

• Side panel, then Agent Builder, then Create New Agent
• Fill in the instructions and pick a model, then Create
• Attach knowledge files only after that, then Save

The TA in Action

Patent Law TA - Week 3: Novelty & Priority

• A student asks whether a claim is anticipated
• The TA refuses to answer first, and asks what the student thinks
• It quotes § 102(a)(1) and points at Claim 1 of the attached patent
• Instructions and knowledge files, doing exactly what you told them to