📍Bangkok, Thailand, 11SEP2025

AI is Peak Mediocrity

Arena di Verona, Italia

If someone is deeply impressed by AI, it immediately tells me something about them. Either that person understands the specific use cases in which AI can be applied effectively, or, they are someone who isn't very good at what they do. This isn't something new; from fashion, to cinema, to architecture, for decades we have witnessed a decline of standards in many facets of life. When I had a few jackets, shirts and slacks tailored recently, I was met in some circles by comments such as "I can't believe you're wearing a blazer". Such comments usually come from fusion-cuisine-eating, hoodie-wearing, Instagram-scrollers.

AI IRL

Recently I noticed a marked increase in the prevalence of AI use amongst my friends and contacts. Even people I considered relatively switched on would copy the responses of an AI chat into our message conversations, or drop them IRL. Some classic examples included:

"ChatGPT said my property should continue to increase in value forever";
"ChatGPT said that at $200k I am severely underpaid"; or,
"My ex will never find anyone like me, probably still regrets losing me, ChatGPT assured me".

It's not that I haven't used AI myself, convincing an LLM to tell you what you want to read is trivial. When I prompted ChatGPT to tell me the likely outcome of the Ukraine war, it gave me the standard EUSSR/NATO line. When I fed it readily available open-source data with respect to the conflict, it switched its opinion like a flag in the wind. When I prompt the models for code, they do great in a narrow context (such as a function or a file) but completely fall apart across a project or on anything remotely difficult.

Why learn anything?

Recently at a table next to me in a Pizzeria in Rome, I heard the familiar twang of the New South Welshman's accent. A boomer couple seated next to us were not enjoying their trip. "Yeeeh narrrr, Rome's orite but nobody speaks English", that was the most pressing concern. Yes, slop culture once again rears its ugly head in the domain of language. Today, even travelling to a far-off land should be "easy" with no real learning done - As Nassim Taleb points out in his book Antifragile, it is when we come up against friction that we learn. Never fear, on your next trip to Italy, the new Apple real-time translation earbuds are ready to relieve you of the need to blurt out a polite "Buona Sera"!

So why might a reverence for AI betray incompetence? To answer this question we need to understand how the LLMs - which most people think of when we talk about AI in the current context - work. AI LLMs are, at their core pattern-matching machines which, noteably, are non-deterministic. In antiquity, AI would have been the village fool, in the modern context it is that lazy, fat, insufferable co-worker who should have been fired long ago.

Everything about AI is mid. Accuracy? Every AI model comes with a warning that everything must be cross-checked and verified - just as you would approach work from your lazy, fat colleague. Conviction? As already discussed, with only the most delicate prompt an AI model will change its mind like a brown-nosing political aide. Reliability? When the AI lies to you, which it so frequently does, it doesn't even tell the same lie twice! This is probably the strongest evidence that AI is sentient - after all, real-world liars can never keep a story straight.

Be the last 20%

As someone who has switched careers, from finance, to technology, my rule of thumb is that it takes only a few years of dedication and hard work to be better than 80% of anyone in practically any given industry. This is because most people don't care, don't have a passion, have other priorities, or are just barely functioning in this artificial and soulless modern economy of fake jobs and fake careers. Given the shortcomings of AI, when one relies on it in their work, they are not relying on some magnificent intelligence, they are relying on distinctly unreliable and possibly factually incorrect information, which only serves to cloud their judgement and corrupt their knowledge base.

Now, invariably some readers - although not you present reader, I know you are one of the smart ones - will scoff and state "if it gets me 80% of the way there that is all I need." In which case, that is fine, this here isn't targeted at you because you don't care about the last 20% anyway. Just remember, my dear friend, that the last 20% is the Italian wine, the German engine, the English suit, and the Russian novel.