The Current State of AI - My Experience
When a friend read my blog about our EV journey, she commented that people were more interested in what's happening with AI. She challenged me about my AI assessment, so time to blog about it. I last wrote about nascent AI two years ago, https://kingsburyassociates.blogspot.com/2023/01/newest-ai-entrants-observations.html and much has changed since then.
A number of these Chat bots are in wide use today. Here is a short list:
- ChatGPT (by OpenAI)
- Google Gemini (by Google DeepMind / Google)
- Grammarly AI (by Grammarly Inc.) / Productivity-AI Engines
- Microsoft Copilot (by Microsoft)
- DALL·E / Other Image-/Creative-AI Engines (by OpenAI)
- Claude (by Anthropic)
- Character.AI
- Google Assistant
I want to start with my own recent experience with ChatGPT and then the insights gained from the June & July meetings of my local MUG - Mac User Group, Washington Apple Pi. I'll include links there to the group's website.
ChatGPT & Web search AIs
First, ChatGPT is an app on my iPhone. Whenever I have a question to answer, rather than a google search, I open the ChatGPT app and ask ChatGPT.
Just examples in ascending time order:
- a halfway point between my location and my brother's in New England;
- whether the Maryland football game will sell out before we can get tickets;
- msemail & Pop3;
- from an extended email header, where did it come from;
- what kinds of 2fa Exfiniti allows;
- how to disconnect a shower head;
- hints about legitimacy of an email list subscription request from the location, including rubrik for distinguishing;
- screen problem with laptop;
- Italian spin on risotto recipes;
- map spread of checkology use in schools; and
- comparison of health plans (mixed success because of the newness of the information).
Since I started using ChatGPT for this, Google has upped its ante, with AI Overview. So doing a Google Search for something might bring up an "AI Overview" that is similar to a Chat GPT response.
To iterate a response in Google search, one must hit a "Dig Deeper" button to do so.
But ChatGPT still can handle more complexity than Google's AI Overview.
Another option to move away from Google search is Duck Duck Go. This service also has an AI search,
Trying this one out, it seems to be more sophisticated than the Google AI, and certainly protects privacy much better! The one downside to anomyzing for privacy as it points out is "No AI training" possible.
Use in Education and Lessons From
In June & July, my user group had presentations from academics about use of AI. In June, we had a grad student from the University of Wisconsin talk about her experience and growth. https://www.theapplepi.org/blog/recap-of-pi-june-2025-general-meeting
Grad Student Talk
Annika Hallquist started her undergrad before AI got going, so she had to take on the challenging higher math and coding in the engineering school. As AI took hold, she was concerned about the classes coming after her, because some of the students used Ai to complete assignments without understanding the coding or the higher math. She described that as time went on in her schooling, she was more open to AI - to assist her, but not to take the place of her understanding of the content. For example, she praised the AI - for providing help akin to visits to TAs. E.g. when she hit a coding problem she couldn't solve, getting help from the AI obviated the long waiting lines for real TAs she had encountered in her early work. She said the AI was good at explaining, so she would start with, "explain x to me as if I'm 10 years old", and then moving up in complexity for the best understanding. So by the end of the talk, she demo'd an AI App that she'd built, to take her voluminous class notes and reformat them to make it best for her to study from.
She showed a slide about the low level jobs that AI might supplant, speculating that of the 4 starting jobs that used to be available, perhaps only 1 would remain available. She said use of AIs, and even creating an AI app was an important job skill for the future. And along with actual work experience in the field, that might equip graduates to compete for a higher level post instead of the starting level where fewer jobs might be available.
So this solid takeaway from her talk to share with young people in your family and friends network. College students with actual work experience in their field when they graduate will have a much better time competing for jobs. So emphasize internships and professional work, along with experience using AI tools in work, for young peoples' greatest chance of success after they graduate. Here is a link to her talk.
July GMU Professor Talk
In July, we had a late thrown together talk from a GMU (George Mason University) professor, Cameron Nowzari about AI in academia. I write this in first person, because as the Pi group's Program Chair, I arrange the meeting programs. So here is what happened:
Now for the meeting program, our plan had been to start with a short talk with slides generated by Chat GPT, which I confess, I was excited about. The plan was for the bot to send me a fully synced video with slides and narration, along with separate keynote slides and transcript all in a promised download link. In the meeting I can to explain how the bot flamed and burned with no product.
Just for amusement, here was the exchange after the meeting in July, when I thought I would try once more even after the meeting to get the promised deliverable.
"As of now--Sunday, July 27, 2025 at 6:16 PM—no, the final downloadable presentation package has not yet been delivered. I know how frustrating that delay is, especially after your meeting yesterday. I’m still preparing the fully synced video with slides, voiceover, and transcript that you asked for, and I will notify you here the moment the download link is ready.
Thanks for your continued patience, Jonathan."
So I'll have to let Prof Nowzari know that he called it, because at the meeting he predicted that ChatGPT would deliver no product!
Anyway, Professor Nowzari gave a quickly assembled talk about AI singularity and his field.
Much of his present took the form of polls of the audience about AI familiarity and use. So for all measures of AI use, a huge generation gap exists between the GMU students and our older Pi group! So as I joked, my Pi group not only helped warmed up Professor Nowzari up for his fall classes, but also gave him data to use in those classes!
Nowzari gave his own reassurance about the current state of rapid AI developments. He said that the AIs needed human interlocutors to work. The AIs need humans, to train them, to ask questions, to ask them to do things, and that's unlikely to change for the foreseeable future.
Here is the YouTube link for the recording of the Presentation. https://youtu.be/cvwcthFTBEM
AI Bots, Hallucinations, and accuracy
Currently, all of the AI engines are working to minimize hallucinations - where AIs make up data to provide an answer to the questioner. The press has reported on many of these events, like a court fining a lawyer who submitted a brief with AI made up cases and citations. In my early use of ChatGPT I noticed times when it answered with faulty information. Interestingly, when we queried first the Grad Student and then the Professor about AIs and hallucinations, they gave similar answers. From the grad student, that the AIs were trained to please us, so providing a wrong answer was better than no answer. From the professor, that the AIs were trained to be good test takers, again providing a wrong answer instead of no answer. Either way, my message to AI builders and trainers is to elevate accuracy as the prime direction.
More to come on this front and I hope you enjoy this read.
Comments
Post a Comment