Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
I'm Kenyan. I Don't Write Like ChatGPT. ChatGPT Writes Like Me - By Marcus Olang
TL;DR
Marcus Olang flips the accusation on its head — his core point is that he doesn’t write like ChatGPT; ChatGPT writes like him, because both were shaped by formal, structured English tied to Kenya’s education system and colonial history.
The so-called 'AI tells' are ordinary Kenyan English-class habits — things internet sleuths flag as robotic, like em dashes, semicolons, transition words such as 'furthermore' and 'moreover,' and clean parallel structure, were explicitly taught for exams like KCPE.
He traces this style to a brutally specific pipeline — the Kenya Certificate of Primary Education composition exam, scored out of 40, rewarded proverb-led openings, 'wow words,' compound-complex sentences, and perfectly scaffolded intros, bodies, and conclusions.
The deeper argument is about bias, not punctuation — AI detectors that judge 'perplexity' and 'burstiness' can misread writers trained to be clear, balanced, and predictable, effectively flagging a whole educational tradition as machine-made.
His sharpest line is that LLMs accidentally reproduced the linguistic ghost of the British Empire — models trained on books, legal documents, academic papers, and encyclopedias end up sounding like the formal Queen’s English many Kenyans were taught to master.
What gets called 'human' is becoming culturally narrow — if authentic writing now means casual errors, American colloquialisms, and informal rhythm, Olang argues that writers from Nairobi, Lagos, Mumbai, and Kingston are being denied the presumption of humanity.
The Breakdown
The post about dashes that finally broke him
Olang opens with the social post that pushed him over the edge: a claim that using a dash is a dead giveaway for ChatGPT, a post that racked up roughly 12.2K likes, 4.2K comments, and 856 reposts. What makes it sting is the reply chorus pointing out the obvious: some people actually had to learn English, and their writing carries that training.
The proposal note that said, basically, sound more human
He recounts getting feedback on a proposal he had worked on for days: strong foundation, but could you rewrite it with a “more human touch” because it sounded like ChatGPT. He plays the moment for comedy — “Human touch. I’ll give you human touch” — before settling into the real frustration: the accusation wasn’t entirely wrong about the style, just completely wrong about where that style came from.
Why his prose sounds structured, balanced, and suspiciously AI-like
Olang admits his writing shares some DNA with LLM output: orderly sentences, transitional phrases, elegant punctuation, and arguments built for maximum clarity. His thesis lands cleanly: he doesn’t write like ChatGPT — ChatGPT, in its globally averaged way, writes like him and millions of others trained in a formal English pipeline.
KCPE, composition drills, and the red-pen algorithms of school
The heart of the video is a vivid tour of Kenyan schooling, especially KCPE and the English composition paper, where a 40-minute sprint could shape your future. He recalls prompts like “A holiday I will never forget,” the mandatory proverb opening, the drilled synonym lists, and the rule that an essay had to be a perfect edifice: introduction, body, conclusion, moral neatly tied off — maybe even with “and that’s when I woke up and realized it was just a dream.”
The colonial backstory behind "good English"
From there he zooms out: this style predates AI by a long shot, because it’s a descendant of the British Empire. The English taught in Kenya was the Queen’s English — the language of administrators, missionaries, law, Shakespeare, and the Bible — and mastering its precision signaled education, class, and access to opportunity.
Why AI ended up sounding like a top composition student
That history sets up his punchline: LLMs are trained on books, encyclopedias, academic papers, and legal documents, so they learn to equate authority with grammatical precision and logical structure. In his words, the machine accidentally replicated “the linguistic ghost of the British Empire,” ending up sounding like a KCPE graduate who scored an A in composition.
The detectors are measuring culture, not just machine-ness
He then explains detector logic in plain language: perplexity measures predictability, while burstiness measures variation in sentence length and rhythm. But Kenyan students were taught to be smooth, balanced, and consistent — in other words, to produce the exact low-perplexity, low-burstiness writing these systems now mark as suspicious.
The real question: whose humanity gets recognized?
Olang closes by widening the frame beyond Kenya to Lagos, Mumbai, Kingston, and Nairobi, asking what happens if “human” writing is defined as informal, error-prone, and American-sounding. His final plea is simple and sharp: before you point and cry AI, consider that you may just be looking at a human shaped by a different education, a different history, and a different standard.
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
