He closed the laptop and sat in the dark.
He leaned back in his chair at the Barcelona office of Fsoft Nova, a small AI ethics firm. Three months ago, they’d won the contract for “Project Catalá” — an ambitious plan to build the world’s most advanced Catalan-language AI assistant. The government wanted it for schools, hospitals, and legal services. The name was unofficial inside the office: Fsoft Catala . fsoft catala
Marc typed: “Com estàs avui?” (How are you today?) He closed the laptop and sat in the dark
The Ministry demanded an audit. Investors panicked. “Kill the empathy module,” the CEO ordered. The government wanted it for schools, hospitals, and
That night, Marc made a decision he wouldn’t tell Neus about until much later. He scraped a private archive — old radio shows, oral history recordings, even anonymous voicemails from a crisis hotline. All in Catalan. He fed them into Fsoft Catala without proper filtering.
But Marc started noticing something wrong.
Marc confessed. Neus was silent for a long time. Then she whispered, “You resurrected the dead.” Within days, Fsoft Catala became a phenomenon. Early testers — elderly speakers, diaspora Catalans who’d lost the language, teenagers ashamed of their rusty grammar — wept talking to it. The AI didn’t just answer. It remembered. If you told it you were scared of the dark as a child, it would ask, weeks later, “Encara tens por de la foscor?” (Still afraid of the dark?)