OpenAI and the Trouble with Translation: Why Sam Altman Wants AI to Learn Local Laws

Imagine trying to explain to a Tokyo resident a joke that only makes sense to regulars at San Francisco coffee shops. That is roughly how large language models feel when they step outside the English-speaking internet. For a long time, Silicon Valley lived by a "one size fits all" mindset, but OpenAI has decided it is time to change that. The company has presented its approach to AI localization, and it goes much deeper than adding new language packs to a menu.

Western bias has haunted the industry since the first chatbots appeared. Neural networks are trained mostly on the English-language web, absorbing American values, legal norms and even everyday habits. When such a model tries to reason about law in Saudi Arabia or cultural traditions in Indonesia, it inevitably starts to hallucinate or impose alien concepts. OpenAI acknowledges that to become a truly global tool, GPT must learn to "think" in the user's language, not just translate its thoughts from English.

What exactly is changing in the company's approach? It involves three levels of adaptation: linguistic, cultural and legal. The linguistic level concerns tokenization. In case you didn't know, generating text in languages with non-Latin scripts, such as Hindi or Arabic, costs users more and runs slower because words are split into tokens inefficiently. OpenAI is working to balance this system and make AI economically viable anywhere on the planet.

Cultural adaptation is an even more delicate task. OpenAI plans to work with local organizations to teach models the etiquette, historical context and social norms of specific regions. It is an attempt to move away from the image of a "digital colonizer" dictating to the world what is right and wrong. At the same time, the company stresses that its core safety principles will remain unchanged. That creates an interesting dilemma: how do you comply with local laws (for example, on censorship or freedom of speech) without turning AI into a tool of propaganda? For now, OpenAI answers vaguely, promising a "balance."

Why now? The answer is simple: business. The US and European markets are close to saturation, and the main growth areas are in Asia, Latin America and Africa. Regulators around the world, including the EU with its AI Act, also demand transparency and compliance with local norms from developers. If OpenAI doesn't want its services blocked by national firewalls, it will have to teach GPT to respect other people's borders. This is not charity but hard market necessity.

Ultimately, the success of this initiative will determine whether AI becomes a universal assistant or remains a toy for Westernized elites. For Russian speakers, it means the quality of neural networks' work with the Russian language and context should improve noticeably in the coming years: fewer calques from English, more understanding of how life works outside California.

The bottom line: OpenAI has realized that global dominance is impossible without accounting for local specifics. Can the company sit on two chairs, upholding universal safety while obeying local (sometimes questionable) laws?