Mid-2026 AI Usage Check

[jon-nyc]

I thought it would be interesting to do periodic surveys of how (or if) we are using AI.

What model or models do you use? If more than one, is there rhyme or reason as to when you choose one over another? Are you paying for any?

Has your use moved beyond the google on steroids? IOW, are you just asking it questions or are you doing more advanced stuff like vibe coding, running multiple agents, etc?

Feel free to add a relevant question I missed.

I'll start:

I have CHatGPT, Claude, Grok, and Gemini apps installed on my phone, all in a neat row on the bottom of my front page. I pay for ChatGPT monthly, I'm finishing up a year's subscription to Grok which I won't renew. I was about to start paying for Claude to get Fable 5 but it was swiped by USG before I could. I imagine I'll be paying for Claude and ChatGPT moving forward.

ChatGPT and Claude get most of my attention (makes sense, they are the two frontier models). Grok I mostly use for the 'explain this tweet' feature. Gemini I use only very occasionally. Very occasionally I'll ask the same question to the four models to see if the answers align (cutting and pasting is trivial, so why not).

I find that once a model gets a lot of history and context with you the switching costs are very real, if not dollar denominated. That keeps me on CHatGPT for a few topics - my work and my health issues - even while I've moved to Claude for much else.

Not done any coding with them yet. I may try to play around with them to either replace or increase the efficiency of some of the bespoke spreadsheets I've been using for many years.

Haven't got any agents working. In that sense I'm still a basic if frequent user.

How about you?

[wtg]

I've used the free version of Claude for reducing imaging reports (MRIs and CT scans) to understandable summaries. Also for interpreting progress reports from physical, occupational, and speech therapists. For both the imaging and the therapy reports I fed in the reports over time and Claude did a good job helping me understand what progress/lack thereof that Mr wtg's therapists are seeing. Saved me a ton of time figuring out all the terminology.

I also used it to help figure out how to use ramps to facilitate wheelchair access to our house. It was lousy at that until I redirected it. And then it was still not very good. It did tell me which manufacturers have the highest quality products. It was probably a toss up whether it actually saved me time or not.

It gave a couple of off-the-wall interpretations about things that happened in therapy. It is much more "emotional" than I expected. I thought it would be more logical and detached.

It did a pretty good job comparing two OTC e-stim devices we were considering for home use to help try to wake up Mr wtg's arm.

I have found its fawning to be annoying. "You're so smart to have figured this out." "Mr wtg is so lucky to have you as an advocate in his corner." I haven't asked it to stop doing this yet, but I'm pretty close.

The mini-AIs that are in Amazon that answer questions about various products are basically useless at this point. And duck.ai (pops up in DuckDuckGo searches) and Leo (part of Brave) also have pretty limited use.

Haven't done anything beyond that.

I feel like I did when I first started using Netscape and AltaVista to access the info on the internet. The information and analysis that are available is amazing. But I'm also a bit troubled by the quality of the results. I still feel like I have to double check a lot of things because it can miss information and its analysis can be flawed. That becomes apparent if you have any background in the topic at hand. The quality of the prompts is definitely a factor; you kind of have to figure out how to use it. If you don't and take it at face value I could see running into problems.

[wtg]

Ha. I just used it to give me the skinny on carpeting. We will be replace our aged plush with something more wheelchair friendly. Claude did a great job.

And I also asked about how to set up and use the NMES gizmo that we bought for e-stim to see if we can wake up the arm. It was very comprehensive.

[ShiroKuro]

For work, I use ChatGPT not to help me, but because I'm studying its output patterns for a research project. I don't use AI in my own writing, or for most teaching things (like grading, no AI for grading).

I might start using Notebook LM a bit more in an upcoming project, I haven't decided yet... and in the fall, I might start using AI (maybe ChatGPT? I want to make some GPTs) for some analysis tasks. Also not decided about that yet. I have a love-hate relationship with AI (like a lot of people...) Also, from all that I done with it, I recognize that it is both incredibly powerful and sometimes so bad that it's laughable. And if you're not already an expert, you are most likely going to miss the big errors and hallucinations.

For personal stuff, I mainly use ChatGPT for health-related stuff, and like @wtg I worry about the accuracy/reliability of the output, so for anything important, I tend to use the AI output as a jumping off point, follow the links provided, and then look for reliable sources once I have a better idea of what to look for. So I feel like that's pretty helpful but I can also imagine how you could get sort of lulled into not paying attention or looking at things with a critical eye, and then internalize something that's actually not correct and potentially harmful, esp. with health stuff.

@wtg said:

I have found its fawning to be annoying. "You're so smart to have figured this out." "Mr wtg is so lucky to have you as an advocate in his corner." I haven't asked it to stop doing this yet, but I'm pretty close.

I agree, that gets annoying very fast. Even just starting with "that's a great question" is tiresome.

[dolmansaxlil]

I have used ChatGPT for unit planning. It is very good at combining two grades together into a coherent plan that addresses both of the curriculum expectations. Gemini and CoPilot were both terrible at this. I have enough curriculum knowledge that I could recognize what it was doing well and give correcting prompts to fix some things. There was no fixing how terribly it botched Indigenous themes in our social studies curriculum - after trying a number of times I gave up and just removed that section from what it gave me and inserted my own section in its place. One thing I did like was I took the ultimate plan and then fed it back into Gemini and CoPilot, asking them to analyze the plan for curriculum coverage and point out weaknesses and any expectations that were missed. These analysis were remarkably similar to my own.

I also use ChatGPT sometimes in my personal life. For example, I need to alter a dress but the construction is unusual and, while I had an approach in mind, I wanted to see if there was another way that might work better. I gave detailed information about how the dress was constructed, explained my problem, and asked for an alteration plan. I was surprised at how well it did at coming up with a workable plan (that was different from the one I had come up with). When I asked some clarifying questions, it did a surprisingly good job of giving me a more detailed set of instructions for one part of the plan. When I asked what it thought about my original plan, it approved it as a simpler fix but cautioned me on one part of the plan based on the fabric the dress is made of. It provided an alternative technique that is common amongst high end dress makers, but one I haven’t used since theatre school so had forgotten about. I am not sure how it would have done with a less skilled prompt. I know the terminology and am well trained in this area - I think that makes a big difference in the quality of the results.

Like others, I have really conflicted feelings about it. I find it useful at times, but am also appalled by the trust people put into it and the number of terrible AI flyers local businesses generate.

[Axtremus]

Little deliberate use of AI, mostly through Google search and whatever Apple devices do on their own.
Want to play with locally inferred AI, but have been too busy.