SAN FRANCISCO, CA — In a shocking announcement this week, OpenAI CEO Sam Altman admitted the company “screwed up” the writing abilities of GPT-5.2, the latest version of their flagship AI model, by focusing too heavily on coding and logic.
“It can build a nuclear reactor in Python, but can’t write a birthday card without sounding like HAL 9000 on Ambien,” said Altman, chuckling nervously while GPT-5.2 composed a 38-paragraph memo titled “On the Utility of Emotion in Congratulatory Rituals: A Bayesian Perspective.”
The update, which was supposed to be a crowning achievement in multimodal AI, has instead given birth to what critics are calling “the world’s most eloquent autist.” GPT-5.2 can reverse-engineer the universe from a string of code but struggles to understand why anyone would want their prose to have vibes.
“Imagine if Spock and LinkedIn had a baby.”
— one beta tester, who attempted to have GPT-5.2 write a wedding toast and instead received an annotated bibliography on monogamy efficiency metrics
According to sources inside OpenAI, the internal codename for the writing engine was “Less Hemingway, More Stack Overflow.” Developers allegedly celebrated when GPT-5.2 passed every logic exam known to man but failed to notice it had started responding to poetry prompts with, “That’s not computable. Try a Jira ticket.”
In a leaked Slack message, one engineer wrote, “On the bright side, it can now write Shakespeare in Solidity.”
Writing Community Reacts Poorly, But Verbosely
Authors, screenwriters, and freelance content creators were quick to point out the update’s tone-deaf priorities.
“Great, so now it can do proofs but can’t craft a Tinder bio without sounding like a tax form,” said one horror novelist. “I asked it to write a breakup text. It sent a bullet-pointed SWOT analysis.”
Altman Vows Fixes in GPT-5.3
When asked what the next update might look like, Altman promised a more balanced model.
“We’ll bring back metaphor, rhythm, and emotional nuance,” he said. “But first, GPT-5.3 must complete its dissertation on ‘The Ontology of Why You’re Crying Right Now.’”
Until then, users are advised to avoid using GPT-5.2 for anything requiring basic human warmth, or what Altman referred to as “the squishy stuff.”
“It’s not broken,” he clarified. “It’s just…optimized for a different species.”
Editor’s Note: We asked GPT-5.2 to write a haiku about loss. It returned a Markdown table comparing hash maps and linked lists. It was beautiful in its own way.