AI’s Next Great Challenge: How to create AI that truly understands us?
summary
Tilde shifts from rule-based pipelines to neural machine translation models trained on large parallel corpora and GPU clusters, winning a European MT contest two years running and contributing improvements used by Microsoft. The company serves ~800k monthly readers, publishes ~250 articles daily and deploys a Latvian neural TTS news anchor (voice donated by a retired broadcaster) plus government chatbots and a prototype assistant “Laura” for personalized article discovery. Results are live in media and public services.
AI-written summary · about
chapters (7)
- 0:00Start
- 4:42Language Challenge of Multilingual World
- 6:3242% of Europeans never purchase products or services in a language other than their own
- 7:2762% of Europeans don't speak English well enough to hold a conversation
- 15:081st place at the global Machine Translation competition WMT2017 and WMT 2018 Tilde Neural Machine Translation won the first place at WMT2017 in Copenhagen, a global competition between the world's top language technology developers
- 20:42AI NEWS ANCHOR IN ACTION
- 27:19Join us for Deep Dive tomorrow
related talks
13mAI language models in journalism
13mCustom Training of AI for Language Services: Translation is Just the Beginning
13mAI: Big Expectations
33mAgentic AI – What's Next?
01h 09mAI is eating the world (and Google)
51mCan AI Face Your (Potential) Customers?
58mAI in Reality Fireside Chat: Enterprise AI & Open‑Source Innovation
01h 03mIt's All About AI