In this article, I speak highly of AI

“How do I disable all WordPress widgets without using plugins?”

“What are Theodor Adorno’s major works — and where should I start reading them?”

“What is the best meditation routine for deep sleep?”

“How do I delete a Docker container from the command line?”

“Are vinegar and baking soda a good combination for household cleaning?”

“In the US, what is the average number of viewers per movie released in a given year? Use a recent year.”

“What does a screen with NCVM IPS technology mean?”

***

Sorry for the random questions. These are some I asked the AI (Duck.ai and Claude) recently. All were answered by the free models offered by the two companies, with varying levels of satisfaction. At the very least, they pointed me toward promising avenues to delve deeper into research, conduct tests, and ultimately solve my problem. (Except for the one about movie box office in the US; it seems data is missing for lower-grossing films.)

In the recent past, I would have turned to a web search engine — DuckDuckGo, which I use by default for more than a decade. They wouldn’t really be questions, because I never got into the habit of “chatting” with the search engine’s text field. Today, I’ve been going straight to commercial chatbots, based on large language models (LLMs). I just ask, the way I would ask a person, though I don’t confuse the two.

And why do searches the old-fashioned way? No matter which search engine you use, the results are similar — if not the same — and almost always poor. Generic, misleading pages with answers buried in wordy texts that are that way “because Google likes it” and because ad revenue takes priority. Content becomes a supporting act — or, as the old saying goes, the tail (ads) wags the dog. Most of the web has gone to rot. Search engines merely reflect this.

For direct answers, an AI returns the summary — that paragraph that would be lost among tons of rambling text and invasive ads on the fifth or sixth link I tried from the search results page.

There are valid concerns about AI as a business. This does not mean, however, that the technology should be discarded. The problem isn’t the technology. It’s the companies and their delusional rhetoric; the business model, the bubble yet to pop; the use of AI as a scapegoat for layoffs and other misguided corporate decisions; the rush to build data centers without considering the side effects, which in many cases are extremely harmful.

Treating AI as a normal technology is an approach that I and others (such as Cory Doctorow) sympathize with. In my case, I came to sympathize with it because its usefulness is undeniable.

AI, on its own, will not end the world nor lead us to the nirvana of the human experience. These apocalyptic or utopian outcomes stem from the business world. It is in the interest of companies in the sector to paint a normal technology as exceptional and bet everything on the idea that AI solves all problems, even if it costs our future and that of the planet. It doesn’t have to be this way.

***

My use of AI/LLMs is limited to trivial questions or those related to my professional activity, and to generating some simple code that I wouldn’t be able to write myself, but at least partially understand what it means.

To date, I’ve completed two major projects with the help of the free version of Claude.

The first was the sliding comments panel on Manual‘s pt_BR edition. It worked, but in the end it felt out of place with the reliable simplicity that characterizes this website.

Earlier this week, out of the blue, I remembered an old goal: to convert my personal blog from Jekyll to Hugo. Both are static site generators, except one is written in Ruby and the other in Go. (These are programming languages.)

I converted it from WordPress to Jekyll in 2017. At the time, I picked a default Jekyll theme and customized it so much that it bore no resemblance to the original. I didn’t really know what I was doing. I only managed it through a lot of trial and error and by reading scattered tutorials on the web. Over time, I became familiar with the code, understood Jekyll better, and learned how to make the edits I wanted.

I never got along with Hugo. The syntax and logic are different from Jekyll’s, harder for my head to wrap around.

Migrating from Jekyll to Hugo was far from a priority. I lacked the time, the desire to learn how another CMS works, and there were absolutely no advantages that justified the effort. Jekyll works fine. I dislike its dependency system (“gems”), the opposite of Hugo, which is just a binary, and… well, that’s it.

Since my blog is one of the simplest and has a well-defined (and limited) scope, I decided to try my luck with AI. I pointed the Jekyll version’s repository to Claude and asked:

I have this blog, built with Jekyll. I’d like to convert it to Hugo — a perfect conversion that preserves the categories (notes, images, texts, etc.) and the layout. In the end, generate a compressed archive containing all the necessary files

We struggled with the filters by post type. After some back-and-forth, it worked. Another step I had to take was asking the AI to give me the Python script to convert the posts themselves from the Jekyll standard (the front-matter) to Hugo’s. Claude was trying to do it itself — that is, on Anthropic’s servers — and kept failing. Locally, all I had to do was call it in the terminal with the command python3 convert_posts.py and watch all 364 posts get converted in less than a second.

You can read our conversation (pt_BR).

With the site up and running, I turned to search engines to learn how to host it on Cloudflare, which, since I uploaded my last site, has removed the simpler “Pages” service to focus everything on “Workers,” which requires some execution code that I had no idea how to create. Hugo’s own website has the recipe.

I haven’t looked closely at the generated code yet. It’s very likely there are areas for improvement. Right off the bat, I noticed that one of the layout files, list.html, was duplicated in two directories. I asked if I could delete one of them. I could.

Otherwise, it worked. And as expected. I base this statement on the time it takes to compile the static site, which dropped by ~77% in the Hugo version. Not surprising (Hugo is faster than Jekyll), but if the reduction hadn’t been so dramatic, it would have been a sign that something was wrong.

***

It’s fair to say that my new site is the result of the collective work of thousands of programmers who have published their work in Jekyll and asked and answered questions on StackOverflow over the years.

The perspective of those troubled by generative AI — developers, publishers, newspapers, and myself, for quite some time — differs from what I believe and strive to practice. My issue isn’t with the violation of intellectual property. It’s with the way these companies have siphoned off humanity’s knowledge to resell it; it’s with the commodification of common knowledge for purely commercial purposes.

If there were an AI made by Wikipedia or Archive.org (public, free, and universally accessible) or if commercial companies didn’t charge for access — to give back the content they’ve amassed from all over the world without authorization or any compensation — I would have zero problem contributing my content to the training of that AI. (Actually, I’ve stopped caring about this… those who visit this blog don’t do so for the same reasons that lead someone to use an AI chatbot.)

Intellectual property and copyright are barriers to knowledge and the intellectual development of each of us, imposed and defended by those with the potential to generate revenue. They create castes and breed inequality. They are pernicious. No wonder it’s a cause championed (when it suits them) by big corporations and, consequently, by people who need to defend themselves against those big corps to survive by doing what they know and love. People who produce intellectual work do need a return, yes, but stop and realize that those who usually enforce property rights are Disney, media conglomerates, and major publishers.

In practical terms, however, intellectual property and vampire-like companies — such as OpenAI, Google, and Anthropic — do exist, which justifies attempts to block their leech-like bots and the multimillion-dollar lawsuits that other massive companies file against them. For my part, in my insignificance, I’ve chosen to give in.

***

PS: A few days ago, I read an article by Scott Jenson proposing a different future in which we deal more with SLMs (small language models), running locally and for specific tasks, especially in a transparent manner. I like that idea.

PS2: For the first time ever, I used an AI editor/reviewer for this article. I chose DeepSeek, which I had never used before. As expected, it found several small typos, but what surprised me was its identification of a flaw in the logic of my reasoning in the section on intellectual property. Impressive.