ChatGPT's late realisation, and what humans will miss most as the AI world progresses
The conversation around AI has been dominated by a simple narrative: speed.
The promise of instant, AI-generated content has led many to believe that traditional roles in creative and technical fields are becoming obsolete. For the language industry, this has often been framed as a "threat" from large language models like ChatGPT.
But a recent development at OpenAI tells a very different story. The company is actively hiring a Localization Manager to "drive localization strategy and execution" with a key focus on "AI-first translation workflows with human-in-the-loop review."
This marks a profound but late-stage realisation: in order to truly scale and deliver quality, even the most sophisticated AI will need human intervention.
The history repeats itself
Yet, ChatGPT’s hiring frenzy isn’t the first time we've seen this play out. A decade ago, the emergence of Google Translate and Microsoft's machine translation solutions sparked similar fears, where everyone thought they would be the end of professional translation services. What happened instead? These tech giants became two of the largest purchasers of localization services in the world.
They quickly learned that translating words is just the beginning. Localizing an experience - whether it’s a website, an advertising campaign, or an educational video - requires a deep understanding of cultural nuances, market specificities, and user behaviour. The same pattern, where big conglomerate LSPs are acquiring smaller ones and buying groups of linguists, is now repeating with generative AI.
The difference in practice: GAI Translate vs. General LLMs
To truly understand why human expertise is critical, consider a typical passage from a corporate legal document. The goal is not just to translate the words, but to convey the precise legal and contextual meaning with zero ambiguity.
|
Original English Source Text "The Assignor warrants that all due diligence materials provided to the Assignee are complete, accurate, and not misleading in any material respect, and that no undisclosed liabilities exist which could impede the transfer of rights as stipulated herein." Translation by a General-purpose LLM (Copilot) "El Cedente garantiza que todos los materiales de diligencia debida proporcionados al Cesionario están completos, son precisos y no engañosos en ningún aspecto material, y que no existen pasivos no revelados que puedan impedir la transferencia de derechos según lo estipulado en este documento." Translation by GAI Translate (with expert review) "El Cedente garantiza que toda la documentación de diligencia debida proporcionada al Cesionario está completa, es precisa y no induce a error en ningún aspecto material, y que no existen pasivos no declarados que puedan impedir la cesión de derechos según se estipula en el presente documento." |
The differences:
- Legal precision. The LLM translates "undisclosed liabilities" as "pasivos no revelados," which is grammatically correct but lacks the legal precision of the option "pasivos no declarados", given in the expert reviewed-version available with GAI.
- Dictum and context. The LLM's output feels slightly academic and less fluid. The expert-reviewed version uses a more natural and legally grounded phrasing, such as "cesión de derechos" (cession of rights) instead of the more literal "transferencia de derechos," reflecting the professional language of legal contracts.
- Format. While both translations preserve the core sentence structure, only the GAI Translate version correctly handles the use of specific, non-translatable loan terms and formal legal register. This is because GAI offers custom glossary, which include automatic translations of context-specific terminologies.
The value of human verification
The core challenge of generative AI is not its ability to generate content, but its tendency to produce inaccuracies and "hallucinations". A large language model has the freedom to move away from the source text and invent information, which is a significant liability in business-critical content.
As one of our linguists, Anna-Christina, puts it:
"What AI doesn’t know - like idiomatic expressions - it will translate extremely literally. And also not consistently. I have worked on texts translated by AI, in which AI randomly changes the translation of the same often repeated phrase. Or it switches between informal and formal language within one text. The larger the text, the more frequent such issues occur."
The most sophisticated tech firms understand that this is where the "human-in-the-loop" becomes indispensable. A skilled linguist and subject matter expert can verify information, ensure cultural relevance, and add the context that a model simply cannot create on its own. This human touch transforms raw AI output into trusted, reliable information.
For the language industry, this shift represents an opportunity, a validation of the core expertise that has always defined our work: quality, accuracy, and cultural integrity.
The future of the language industry
The future of language is not a zero-sum game between humans and machines. It’s a strategic partnership where AI handles the scale and speed, and human experts provide the crucial layer of verification, quality control, and cultural and contextual insight. This is GAI Translate's core philosophy: an ecosystem where technology generates value to businesses by empowering linguists, rather than replacing them.
The growing demand for human-in-the-loop solutions from the very companies that created the most powerful AI models confirms a shocking truth: in a world of endless information, trust then becomes the new currency. The companies that will thrive are those that invest in a process that goes beyond simple automation to deliver genuine, human-verified quality.
Discover the world's first secure AI translation tool with one-click expert review here.
