Day: February 4, 2026

I am pretty skeptical about all this Artificial Intelligence headlines in the news, maybe because I know from my study of machine learning that most of the large language models are just cute parlor tricks, that often produce overly soppy essays, half answers and broken code. But in many cases, it actually is pretty convenient way to get answers to questions on the internet, especially simple problems.

Often I forget how to implement a piece of code I’m writing, and I will put out a quick question to the Meta AI or nowadays more and more the Google AI website and it will spit out a mostly working solution to my problems. Or I just forget a specific flag or way of doing bash scripting, and rather then browse the man page or some internet forum, Google AI can put out a reasonably correct answer, especially when the question is fairly black and white.

Likewise, I have asked Google AI several times for information on various Ford SuperDuty trucks, what the different packages me and used it to explore some of my own thoughts on buying a big truck. While AI is overly deferential and arguably not questioning my own line of thoughts enough, I do see value in having a third party explore the pros and cons of various options. It’s a good way to cut through crap and get you answers, though it’s good to verify as AI as the disclaimers often note, makes mistakes.

I still have my doubts about AI being a know-all remote service covering the whole world’s vast knowledge as synthentized through the internet, when most knowledge really is localized. I don’t like being dependent on good internet access or a remote service to get answers. Indeed, I think many of the benefits of AI will become more practical when models are run locally, and AI becomes what it is best – not a generator but an indexer of content. AI makes a better search engine, it does not make a great essay or email drafter, and while the artwork it generates is interesting, I am not convinced it will ever replace humans. AI is a great starting point, a great reference, but ultimately there is something about the human brain and it’s creativity no model trained on vast data can completely replace.