\n\tBy MARTY LEVINE<\/p>\n
\n\tPlaying in the \u201cAI Sandbox,\u201d a new hands-on workshop offered by the\u00a0Faculty and Staff Development Program<\/a>, demonstrated how to get AI to respond with a variety of solutions to basic writing tasks \u2013 and how to determine whether AI is actually useful to you.<\/p>\n \n\tThe presenter, J. D. Wright (English faculty member and consultant to the University Center for Teaching and Learning), offered those in attendance a chance to practice creating suitable prompts, refining them to ask AI for different iterations of the desired document, and assessing whether the experience was helpful.<\/p>\n \n\tThe longer the prompt, and the more detailed, the more complete and accurate the output, Wright said. \u201cInputs that go on for pages and pages, quite literally…. that gets you better and better results.<\/p>\n \n\t\u201cDon’t settle for the first output,\u201d he added \u2013 a habit too often of students, he has observed. \u201cThey take for granted that it is the right answer, that there was no better answer to be gleaned. Have a conversation\u201d with AI. \u201cIt works very much like a text exchange with a friend or a colleague.\u201d Think of AI as a collaborator, Wright said.<\/p>\n \n\tStudents in the class were asked to submit a prompt to Pitt\u2019s Claude AI, or another AI of their choice, to create an invitation to an event in their department, along the lines of \u201cDraft a social media post inviting people to a guest lecture about e-mail etiquette by Prof. Lessing at 10:00 am on Friday in the conference room.\u201d<\/p>\n \n\tOne participant volunteered that Claude \u201cdid exactly what I wanted it to do and made assumptions on its own that I don’t disagree with\u201d such as adding an introduction, leaving spaces for more details and providing a link to RSVP by a certain date. Another participant, who asked AI to make the invitation \u201cfun,\u201d was happy when \u201cit made a really cute e-mail \u2013 with vivid colors.\u201d<\/p>\n \n\tYet another, who said his event involved \u201ca contentious issue,\u201d found that AI suggested a number of arguments that might arise when confronting this issue, adding \u201cfall-back positions\u201d for event organizers to consider.\u00a0<\/p>\n \n\tWright kept adding new suggestions to make the prompts more specific, such as telling Claude what tone was desired for the invitation or what attendees would be required to do in preparation for the event. Some participants appreciated the new tones AI produced, but found AI was also making assumptions that their prompt had not specified.<\/p>\n \n\tAI is not a correct answer machine, Wright emphasized. Rather, it is a tool for generating options. He recalled asking AI to come up with 100 titles for a paper he had written, \u201cand it takes no more time to come up with 100 than it does to come up with five,\u201d he said. Asking for 100 \u201cincreases the odds that it might stumble\u201d upon the one you like, Wright pointed out. Indeed, most of the paper titles it produced were not suitable, but a few were quite good, he said.<\/p>\n \n\tWas your AI experience useful? Wright recommended asking: \u201cDid this save you time or give you a better product then you would have come up with on your own?\u00a0<\/p>\n \n\t\u201cOne of the coolest capabilities of AI is the capacity to be its own best user guide,\u201d he said. \u201cAsk Claude what you’re overlooking. Ask Claude what next questions you should pose. Ask Claude to predict ways that your output could be misinterpreted or unintentionally give offense. Ask Claude for a recommended workflow for doing this task next time.\u201d<\/p>\n \n\tAI users should also consider which tasks they would love to place upon the digital shoulders of AI and what tasks they enjoy and would prefer to do on their own. But even for the latter tasks, he said: \u201cWhat parts of the task could Al streamline? How could Al enhance your enjoyment of the task?\u201d<\/p>\n \n\tOne participant said of her AI experiences to date: \u201cI like it, but I don’t always trust it. I don’t know if it is saving me time,” because she still needs to check its work.<\/p>\n \n\t\u201cI rarely use AI to create a document,\u201d Wright allowed. But he recommends that people don’t hesitate to use AI to create a first draft \u201cyou can tinker with,\u201d he said. Facing the blank page with anxiety? \u201cI never have to deal with that.\u201d<\/p>\n \n\t\u00a0<\/p>\n \n\tMarty Levine is a staff writer for the University Times.\u00a0Reach him at\u00a0martyl@pitt.edu<\/a>\u00a0or\u00a0412-758-4859.<\/p>\n \n\tHave a story idea or news to share?\u00a0Share<\/a>\u00a0it with the University Times.<\/p>\n