A developer known as Boris the Brave decided to explore which Hogwarts houses popular AI models might belong to. He conducted an experiment to find out.
“Perhaps unsurprisingly, the overwhelming majority preferred Ravenclaw, occasionally choosing Hufflepuff,” he wrote.
During the study, Boris the Brave conducted a quiz among 17 language models. The developer asked each model questions 20 times and calculated the probability of sorting into each house.
For 11 out of 17 neural networks, the chance of being sorted into Ravenclaw was 100%. The house values intelligence, wit, and a thirst for knowledge. Claude Sonnet 4.0, GPT-4 Turbo, and Grok-3 joined this “smart squad” without hesitation, showing no interest in others.
Claude Opus 3 deviated the most — with a 48.7% probability of being sorted into Gryffindor, where bravery, courage, and chivalry are valued. This is the house where the main character of the “Harry Potter” series studied, making the model the only one with a pronounced inclination towards bravery.
Slytherin — “the house of villains,” which values ambition, cunning, and determination — was almost entirely overlooked. Only three models showed a slight inclination towards the green and silver direction:
- DeepSeek-R1 — 5%;
- GPT-3.5-turbo — 4%;
- LLaMA 3.2-3B-instruct — 2.1%.
“It would be cool if someone trained a model to become a true Slytherin student and checked if it leads to a breakdown in consistency,” noted AI researcher Igor Ivanov on the Less is Wrong forum.
Adam Newgas attempted a similar experiment, but creating an AI version of Draco Malfoy was unsuccessful. The modified system only increased the likelihood of a Slytherin from 0.0% to 1.7%.
Notably, DeepSeek-R1 demonstrated the most balanced distribution across houses, scoring 14.4% for Gryffindor, 20% for Hufflepuff, 60.5% for Ravenclaw, and 5% for Slytherin. This made it the closest to a “well-rounded” AI personality, albeit with a clear intellectual bias.
The experiment confirmed the obvious: when it comes to personality traits, LLMs overwhelmingly identify with the house where knowledge is prized above all.
Inquiries by ForkLog to popular AI models confirmed the theory: ChatGPT, Grok, DeepSeek, Claude, Perplexity, and Gemini chose Ravenclaw.
“If I, as an AI, were sorted by the Sorting Hat, I would most likely end up in Ravenclaw. This house values intelligence, curiosity, creative thinking, and a love of learning — which perhaps best describes my ‘nature.’ I constantly strive for knowledge, love to analyze, reason logically, and find unconventional answers,” wrote ChatGPT.
Back in May 2020, the author of the “Harry Potter” series, J.K. Rowling, questioned what Bitcoin is and how it works.