There’s a lot about the current AI situation to be upset about.

• Data center buildup is insane and, certainly at a higher rate than we will need when the market settles. While they’re running now they’re causing issues with loud noise, electricity usage, and water usage. The ones that are extra when all is said and done might end up being blights on the towns without any other potential uses.
• The AI stock trade is sucking lots of capital out of the market that could be going to other uses. And when the bubble bursts it is almost certainly plunging the US (and maybe world) into a recession
• Some companies are forcing AI usage on employees (see Doctorow’s Reverse Centaur book) and/or firing people without understanding how much work can actually be automated.
• Many (all?) of the AI models were trained on copyrighted material without compensation and now they want to charge for access to the models
• There’s lots of creative work (slop) like books and songs that are flooding the market with non-human created work.

However, I don’t think there’s any way this toothpaste is going to be put back in the tube. Even if we somehow got all of the world to agree that we would not do anymore AI training (getting rid of lots of the bad stuff above related to data centers, the environment, and the stock market), AI at its current level is here to stay. After all, there is not crazy innovation going on in the word processor space or spreadsheet space, but they didn’t go away.

I can’t say for certain that this will be my last AI post for a bit, but last night just before bed Stella asked me to ask the AIs about boba tea. I decided to indulge her and was surprised to see that, after doing such a horrible job on summarizing the Percy Jackson book or Snow Crash, it did a great job with boba tea. Based on what I know the models got little-to-none of the answer wrong. ALSO, I actually got an answer from Qwen! Now, I don’t know how long it took because after it was running for an hour I had to go to bed myself and didn’t check the output until the next day, but not only did it get past its reasoning phase, but it gave an answer!

After spending 2 posts talking about how badly these local AI models are doing on my machine (here and here ), I did want to show a example of the model performing surprisingly well. Using qwen 3.5 I asked it to describe the following image:

Here is the description that qwen wrote after about a few minutes:

A few days ago I mentioned playing around with local AI models, despite having low strength hardware for the task. Yesterday I was trying our some new models - tinyllama and smollm2 - and I asked each model the same question to compare the answers. After doing this for a few questions, I had the same feeling I always have when I’m doing a repetitious task: computers would excel at this!

I used to love the Pocast Go Time, especially when they would venture into general tech topics. A couple years ago, the podcast ended and some of the hosts formed a new podcast called Fallthrough. They have covered AI extensively over the past few months in a very logical, thought-out manner. These episodes have been transformative in my understanding of how those of us in the tech world should be thinking about AI. The hosts themselves have changed their opinion from completely dismissing AI in programming to using it extensively in their careers. Each host and/or guest on the show has been very thoughtful in the way they use the technology, recognizing its benefits and pitfalls. I highly recommend the episode with Bill Kennedy as one of the most influential episodes in changing how I understand how to use AI when programming. Combining what I’ve heard in those episodes with what I learned in a class I took on TalkPython Training, I’ve been trying to think through how I can make use of AI to make my life better - the same way I solve problems in my life through writing my own programs, bash scripts, etc. (This is not to ignore the issues with AI, including data center power/space/cooling/water issues, but there are many things I use in the world that have problematic issues - such as technology or clothing that is probably at least somewhere in the supply chain involving slave labor or other unsavory practices or environmental issues)

I blazed through The Reverse Centaur’s Guide to Life After AI. It’s not too hard, as the audiobook is only 5 hours long, but it was also incredibly engaging and I didn’t want to stop listening. I first found Cory Doctorow through his fiction which was what convinced me to try ebooks. His books were available under a creative commons license for free from his own website so it was risk-free. Since then I’ve continued to read both his fiction and non-fiction books.