I Really Don't Get the AI Craze
February 17, 2025
I spend all day writing software, and I still don’t get the whole AI craze happening currently.
Anecdotally, whenever I’ve used ChatGPT for work, I would estimate it to have about a 50% accuracy when I ask it a question. Most of my queries involve looking up API documentation that’s hard to find, or diagnosing obscure errors; things that are extremely niche and difficult to find using a regular search engine. This is something I’d expect an LLM to do well: it should use the vast background of its training data to find me an answer, if one exists. About half of the time, it does. The other half of the time, it just makes something up, which is not helpful when you’re in the weeds on a production issue. Admittedly, ChatGPT has bailed me out in the past, but other times it’s just added to the background noise.
I never ask it to write code for me, I never use the code it produces when I don’t ask, and I absolutely never give it any personal information or sensitive data like JWT tokens, PII, unique IDs, or anything that shouldn’t be posted on a billboard. You shouldn’t either. These LLMs are not your personal buddies, they’re computer programs running on someone else’s server, who may not have your best interests in mind. This may mean that my use case of these things is very limited compared to other people. I only really ask these things for help when I’ve run out of ideas, and I’m actually able to ask it a question without any opsec concerns.
So, given this limited use case, my overall rating of ChatGPT is: meh. It’s neat, and only sometimes helpful. Definitely not earth shattering for me, and I’d be able to get along just fine without it.
As an additional anecdote, I don’t think ChatGPT presents a competitive edge for software engineers. I work with some really smart and hardworking guys. In my opinion, the thing that best predicts success amongst the engineers I know isn’t IQ or background knowledge, it’s the ability to grind a problem and not give up. Brains are important, but work ethic is moreso. Using an LLM doesn’t make you a 10x Engineer immediately. It often just generates a bunch of code that you have to spend extra time interpreting, customizing, and bugfixing to work for your specific problem. It’s not really a gain of efficiency. I’m also very particular about the way my code looks, how I comment it, and my overall “style” of programming. Beyond aesthetic value, these aspects of personalization actually contribute to your efficiency as an engineer, the same way a craftsman has honed his particular style and techniques over many hours on the bench. Programming is not a traditional “white collar” job; it is more akin to a trade. Trades pay by experience, experience is a form of wealth, and wealth obtained quickly is quickly lost (Proverbs 13:11). ChatGPT is not an exception here.
But enough about programming. People use LLMs for all sorts of nonsense beyond writing code. Businesspeople use it to make presentations. Students use it to do their homework. Teachers use it to write full lesson plans and assignments. Some use it to dialogue with historical figures, or summarize large manuscripts or concepts. Other AIs write songs and poems, or develop entire web applications. Some blogs are completely AI generated. The question for me is, why?
These things seem to be symptomatic of the modern mind’s preeminent desire for speed and utility above all else. “We want it now, damn the quality, just let the machines do the work.” Now, machines are wonderful, but machine was made for body and mind, not body and mind for machine. What the technologists of the early 20th century envisioned were tools to enhance man’s faculties, not replace them (or worse, take advantage of them). They wanted “a bicycle for the mind”, but we have made our minds bicycles, pedaled by endless feeds, algorithms, automations, and content, content, content…
Now, you don’t even need to think, you can just ask ChatGPT.
If the 20th Century made men without chests, then the 21st Century will make men without brains. It may also make them without hearts, hands, feet, eyes and ears, and souls? Wait, has it already? Is this progress? Was the end goal of progress always the abolition of ourselves?
And don’t get me started on the new Chinese AI, DeepSeek. Just a hunch: this is a total intelligence psyop to collect American data for an American political. Of course every American soydev and bugman is going to buy into it. “Ehrrrrm, but it scored an 89.1 on MMLU-Redux versus GPT-4o 88.0, you should really consider it…”
Don’t talk to the AI, people.
Support Your Friendly Neighborhood Paleocon
Bitcoin
12VJL3YtywTEDVufkk3kbV1AUuxaQPpRz5
Monero
44V6YEA8vhP6hbyAZfJfHpaQZ5MubahHJDeyKohGxSo4FHvZ1T2WdFzM5A6PsQt9SWL4Y5gvPnJQnaKxW2iWHrFsP7J1KD5