My journey to LLM hater

At the beginning of copilot days I looked for an ai autocomplete that cared about licensing and only used permissive software to train from, not big tech, gdpr approved. At the time I used it sparingly only using it to generate the next 3 lines of code. This was the time that general usage only required double digit token cost.

Science vs Slop factory

While aigen images were starting to be adopted by unethical game studios, the advancements in medicine was a good enough reason to use AI sparingly as a autocomplete or search tool. But the industry motivations were starting to slip threw the cracks: job losses linked to LLM's started surfacing in translators, writers, background artists.

IntelliSense replacement

But the longer I used it the more I limited its functionality. It got to a point that I missed intellisence as it would often give a list of methods with some description. That would have been fine if the autocomplete was a different key or displayed the intellisense elsewhere but alas.

Scrape and DDOS

I thought I was safe with a smaller company that advertised itself by listing the open source permissive licenses it used to train its modal, I thought it would be a stepping stone to local modals that help me understand codebases, a static tool. I expected the reports of web scraping to subside after the modals became more specific, weren't these companies promising net zero?

I emailed my beloved small LLM company to query their carbon emissions and if they offset or calculated the tCO2e after all most hosting companies do that. Their response was no, but they use Google Cloud as hosting company For-Fucksake! Later in they pivoted to enterprise only and refunded my one and final year subscription.

Agent Smith

Around that time agents came a long. That was the moment I realised AI in tech was not a new tool but an attempt to remove at least 3/4 of devs & leave the leftover 1/4 to work alone talking to bots. The first thing I did was read the Communist Manifesto ☭.

The Commune

People around me were using the most obese, corporate, pirated modals out there (Perplexity, Claud, GPT5) for: submitting research, calculating, planing their holidays, generating entire websites. Without even looking at the sources, the calculations. Nor was there any consideration on the other uses their subscribed ai corp was selling its modal for: police thought crime, hidden sexist/racist preferences in cv filtering, or the under-reported carbon emissions. If it came up the response was *shrug* everyone uses it, don't you?

The fediverse had better taste, people shared ways to alert me if a Github project was written by agents, boosted no ai stickers, shared LLM free alternatives. It radicalized me by the morals I thought I had but forgot to fully utilize.