Could the road to artificial general intelligence run through something as everyday as translation? That now looks a lot less far-fetched. A long-running trend tracked by language company Translated suggests machines are closing the gap with human translators faster than many people realize. <\/p>\n\n\n\n
Its key metric, called “Time to Edit,” measures how long professional editors need to fix AI-generated translations. In 2015, that took about 3.5 seconds per word.<\/p>\n\n\n\n
Today, it is closer to 2 seconds. Human translators, by comparison, spend roughly 1 second per word reviewing another person\u2019s work. If that curve keeps moving the same way, the company says language “singularity” may come within just a few years.<\/p>\n\n\n\n
Why does that matter beyond the tech world? Because translation sits quietly inside global business, customer service<\/a>, science, and public communication. <\/p>\n\n\n\n In practical terms, better AI translation could help companies sell in more markets, help researchers share findings faster, and help governments get urgent information across borders with fewer delays.<\/p>\n\n\n\n Translated says its systems have already been trained and refined on billions of edits made by 136,000 professional freelancers, and by late 2025 Lara had processed 3.14 trillion words at production scale. That is not a lab toy anymore. It is infrastructure.<\/p>\n\n\n\n But there is another side to this story, and it is one readers should keep in mind. The race toward better AI is also a race toward more computing power<\/a>. <\/p>\n\n\n\n Translated said Lara was trained using 1.2 million GPU hours, and in late 2024 the company said it planned to use 20 million GPU hours in 2025 to move closer to language singularity.<\/p>\n\n\n\n That kind of scale has a real-world footprint. The electric bill<\/a> matters. So does the environmental bill. In fact, Translated has said AI tasks once made up about two thirds of its total energy consumption, which helps explain why it restored a hydroelectric plant in Italy to support what it called “clean AI”.<\/p>\n\n\n\n Still, near-human translation is not the same thing as true AGI. That is where the hype needs a little cooling off. Even Translated\u2019s more recent public messaging has stressed that today\u2019s models<\/a> can still struggle when precision, accountability, and trust really matter. <\/p>\n\n\n\nAlso Read: Seattle installs smart metal on a bridge that bends with earthquakes, does not break, and returns to its place on its own<\/a><\/h4>\n<\/div>\n<\/div><\/div>\n<\/div>\n\n\n\n
Also Read: Thieves alter and cash a $28,000 check, and the bank refuses to return the money to the owners<\/a><\/h4>\n<\/div>\n<\/div><\/div>\n<\/div>\n\n\n\n