Artificial Intelligence ( Part 4 )
Had a conversation this morning that reminded me about this draft on artificial intelligence. The last couple of years have been a doozy with the machines… you know – those little ubiquitous devices that are now embedded in everything. Our dishwasher broke not too long ago and it was because of a faulty computer board. Needless to say, even though I have a love-hate relationship, I use Artificial Intelligence agents and models on an almost daily basis now.
The majority of today was spent evaluating the new GTP-5 model from OpenAI1. I’ve now gotten to where AI is like a pair coding partner. I still don’t like the term vibe coding, but I do use a microphone regularly to interact so it is feeling a bit more like vibe coding. I switched back to using Claude Sonnet 3.72 after several hours of OpenAI’s new model mainly because it’s faster. The better half walked in on me having a full on voice conversation working on my current project today and she just stood there and watched me work for a bit. It’s definetely something else. I use em in agent mode now and wire up the browser reloads on file changes so to her, it just looks like I say something and it flips through the files, makes, edits, and viola the app changes.
If you’re following along, my last essays on AI3,4,5 mainly revolved around finding it useful and applying it using some tools to publish a personal AI assistant that uses all of my own notes. I left myself with the task of digging in to learn more. AI has now because indispensable in my work and this should be a subtle cue that it’ll also be for your work too, whatever it may be. I’ve leaned enough now to have a more robust understanding but I don’t yet feel nearly proficient in the underpinnings because so much of the work has already been done for me by other’s contributions. I inevitably run into research papers that are way over my head and I started in on a course from the HAN ( Hardware Accelerated Neural-nets ) Lab at MIT called TinyML and Efficient Deep Learning Computing6. I’ve as of yet to complete it because of ya know… life and work.
At this point, I know enough to understand exactly how potent the technology will be in the near future but I’d still consider myself a hobbyist as I don’t have near the amount of ability of knowledge to advise. I’ve started experimenting with more commercial applications even though I’ve previously turned away requests mainly because I don’t want the liability of trying to maintain a complex system that I don’t completely understand. I’m stuck way down at the bottom of the stack doing fine turning because I’m dependent on the work of others up top with neural networks, machine learning, large language models, tokenization, transformers, and inference. Regardless, I have an enough knowledge now to fairly confidently explain it to others. There are a couple key elements to understand in order to know how these Artificial Intelligence agents work. I think that the key word to understand is probabilistic7. Probabilistic in software or elsewhere is really just the mathematical relationships of probability. I think It’s important to understand that probabilistic systems acknowledge uncertainty.
Folks, or at least the humor bent one’s I follow on social, like to point out AI halucinations and other errors. And as far as ‘vibe coding’ goes, I can’t count the number of times I’ve experienced a mess of code using an assortment of agents and models. I’m regularly editing out errors on the fly. What I have found though is that AI agents are extremely helpful in my work to the point of no return. I’m still very skeptical of the term intellegence as I’ve mentioned serveral times in this series of essays, but I suppose that’s just semantics. There is no doubt that aside from my hardware it’s the most valuable tool in my toolbox.
Artificial Intellegence will be everywhere and it will be useful in many cases. Like anything powerful, the negative aspects will be equally dangerous especially when it comes to generating false information and media. The same sort of augmented intellect we got when the internet landed in our palms is about to get a serious boost and It’ll certainly have consequences. We’ll be saying the same thing teachers used to say about Wikipedia… “you can’t trust what ( Wikipedia ) your AI agent told you”. I’m not going into it with the same idealism I originally had for the internet, but I’m also not throwing the blinders on either and hedging my bets. I feel much more confident now that it’s never going to take my job especially give the fact that I’ve learned to use it effectively. I’ll just be overseeing it like it’s my paid employee because I’m just making sure it’s a little further down the food chain than I am. I’m comfortable enough to accept legal or medical advice from artificial intelligence agents, but I’ll certainly be confirming it through a doc or lawyer with some experience.
- GPT-5 – OpenAI – https://openai.com/gpt-5/
- Claude Sonnet – https://www.anthropic.com/claude
- Artificial Intelligence ( Part 3 ) – https://davidawindham.com/artificial-intelligence-part-3/
- Artificial Intelligence ( Part 2 ) – https://davidawindham.com/artificial-intelligence-2/
- Artificial Intelligence – https://davidawindham.com/artificial-intelligence/
- TinyML and Efficient Deep Learning Computing – https://hanlab.mit.edu/courses/2024-fall-65940
- Probabilistic – https://en.wikipedia.org/wiki/Probabilistic_method