I’m giving a talk on generative AI at the ACES Conference next week, and I’m curious how people here are using generative AI productively. I’m thinking about things like ChatGPT and other chatbots, DALL-E and Stable Diffusion and other image creation systems, and systems like ChatPDF that can summarize uploaded PDFs.
This is NOT an invitation to opine in general about generative AI. I know you have lots of opinions—I share many of them. But that’s not the goal of this thread. I’m only interested in what you’re actually using generative AI for in the real world now, and I will delete all posts that aren’t some form of “I’m using X to do Y.”
Using it in the photoshop beta…not to generate or materially change images…but for things like making the border slightly more if the original photo was framed too close. It works to add a little more background.
I use ChatGPT to find facts that I have problems finding with Google search. Examples:
What is the sink rate for the Airflo Kelly Galloup Streamer Max Long 200 grain? (A fly fishing line I bought 10 years ago).
What is Active 3-way describing about a single speaker?
What application is reporting “Detected unsafe path transition” on Redhat?
It is important to point the solar panel directly at the sun. What is the performance loss at 5, 10, 20 degrees off direct?
Making very simple scripts:
Make a python script that takes hours:minutes:seconds and converts to seconds.
I have tried 5 times to use it to solve more complex programming tasks; it has been unsuccessful.
I use AI in Photoshop for masking, changing the background, erasing things, expanding an image etc. Sometimes the results are good enough to use as they are, usually they serve as an advanced starting point for manual touch-up.
Not being an artist, I can use Photoshop and ChatGPT to create images from scratch. Sometimes the result is actually usable as is. More often they’re very useful to show the client options when coming up with roughs for a project. They make it easier to pinpoint what the client likes/doesn’t like before briefing/contracting an illustrator or, photographer thus reducing cost for the client and frustration for everyone.
I once asked ChatGPT to use TKinter to build an example of a dual-window user interface in python. (I wasn’t sure it was possible to make a dual-window interface with TKinter, because TKinter documentation is mostly about obsolete versions.) The code it produced was close enough that I could get the example to work, validating that it was possible.
Subsequently I decided that a dual-window interface was not the best for the simple script I wanted, but the insight I gained was useful.
We use the generative feature of Photoshop quite a bit - typically for expanding/extending images which don’t fit the proportions we need. It’s also great for removing areas we might need to simplify to hold text etc. It’s actually quite remarkable what it can do (although it does occasionally ‘miss’).
The new version of FireFly creates images from nothing and I’m sure this will be handy when we don’t have an appropriate stock photo to use…
I have very little current need for ChatGPT but I suspect it will come.
Working up a set of classes on generative AI for my students, seeking to identify ethically sound usage such as the following prompt designed to aid a literature review, uses a prompt with seeds, guiderails and clear tasks.
Roles: You are an expert librarian at a research university specialising in humanities research. I am a research student. Task: Create a reading list providing a basis in the most influential research in contemporary Irish literature, specifically work by women authors. Instruction: Provide a list of most-cited, peer reviewed articles that will help me to understand the topic. Cite your sources and provide links. Only include references, together with URLs from Google Scholar. Detail how many citations each resource has. Only include resources with over 500 citations. For each recommended reading, provide a 5 line summary of its content, clarifying its contribution to the discipline.
There’s a lot of activity in academia on this front, I’ve been gathering other prompts, students are already using this technology, staff have to provide guidance. Our STEM staff have few issues, my Arts and Humanities colleagues have huge issues with it. Curious, Adam, as to your outcome…
I do software implementations and i work with clients for three to six months. We have weekly Teams calls to discuss requirements, updates and progress. I use fireflies.ai to get a text transcript but with each call around 9000 words, I need a more condensed version to scan to remind myself what was discussed on previous calls.
I have a very tediously created prompt that summarized the pdf transcript into a table of info useful to me i use GPT 3.5 using a combo of Keyboard Maestro macros and Apple Shortcuts.
It’s easy to tell if the summary contains any made up info since i run this right after the call.
Besides catching spelling and punctuation issues, Grammarly suggests alternative wording. I use the suggestions a fair portion of the time. Hopefully, that has improved my postings to Tidbits Talk and other forums.
Zoom recently added the option to create an AI summary of a Zoom session. I’ve only used it once, but I found the summary quite useful.
I use ChatGPT when I have a question that could likely be answered with a summary of currently available content, but which is difficult to answer unless you just happen to craft the perfect Google search prompt that results in the perfect article on the topic.
For example, I’m getting cataract surgery next month. I’m trying to choose between the fully-covered traditional (scalpel) method, and the newer laser method, which would involve significant out-of-pocket expense. I asked ChatGPT to summarize the differences between the two procedures, and found the result extremely helpful.
On the other hand, asking it for help with specific programming problems has been hit-and-miss, mostly miss. It can usually set me down the right path, but the specific solutions offered are frequently just wrong.
I think academia will find more uses of chatbots than most because so many of the questions are collecting and summarizing and evaluating data, sometimes very large sets of data. Big business is already doing this, I’ve heard.
My talk is to Apple consultants, so for them, this sort of analysis and summarization isn’t as helpful.
I’ve used Zoom’s meeting summarization and it’s… OK? I mean, it’s better than nothing, but there are some howlers when I’m reading it afterward. A friend said he uses Microsoft CoPilot’s summarization in Teams, and it has been a life-changer because he no longer has to take notes. I don’t know if that’s because it’s better than Zoom or because it’s sufficient for his needs.
We have started a project here with a postdoc working on how to use Google NotebookLM to summarize large bodies of scholarly work and journal articles so we can ask it specific scientific questions that it answers with a summary based on the prior work, and most importantly, the exact sources for its answers.
I’m a free-lance translator (humanities, Japanese to English), and I’ve been experimenting with generative AI to produce initial draft translations. The translations produced by ChatGPT 3.5 were pretty much unusable, but Gemini AI is surprisingly good, though only for an initial draft, and I feel that it has improved my productivity.
I’m the treasurer for a sporting club. We receive PDFs listing members who attended an event, and the charges they incurred over each day of a multiple day event. I have asked ChatGPT to collate the charges for each member, and to output the data as a list so I can on-charge the amounts.
Initially there were a few issues, but I have now changed my instructions to ChatGPT to require it to repeat its analysis if the sum of the charges doesn’t reconcile with the list it outputs.