Building an AI-Powered Organization
summary
DeepL becomes vertically integrated, trains its own LLMs and builds on-prem compute with DeepL Mercury (an 8100 superpod in Sweden) to move from translation to generative assistants. Decision in 2022 reassigns 80% of research to LLMs; April 2024 launches DeepL Pro and ships Clarify, a feature that flags uncertain translations and asks users for missing context. Result: higher-quality, more assistive products but initial tradeoffs in latency that drive follow‑on engineering and a new DGX B200 superpod.
AI-written summary · about
chapters (17)
- 0:00Introduction and speaker welcome
- 1:03Key challenges in building AI companies
- 2:54Why DeepL was the right choice for impact
- 7:28Vertical integration as strategic advantage
- 8:11Mission-driven product development approach
- 12:04Navigating AI trends and long-term focus
- 13:32Embracing failure as a learning tool
- 14:15Clarify and human-AI collaboration
- 17:47Progress as layered S-curves of innovation
- 18:51Transition to LLMs in 2022
- 20:18Building Deep Mercury supercomputer
- 21:42Impact of LLMs on product development
- 22:25Releasing DeepL Pro and NextG models
- 24:12Efficiency breakthroughs in LLMs
- 24:57DeepL Clarify and future plans
- 25:58Next-gen AI factory and Blackwell architecture
- 26:42Q&A on language insights and surprises
related talks
35mDeepL: A Product Culture that Creates Top-Notch AI Products
28mHow to Compete with OpenAI and Google
39mLeveraging AI and Emerging Technologies for Competitive Advantage
19mComputers are still dumb: bringing your AI magic to enterprises
24mNew AI Business Cases
40mBeyond the Hype: Realistic Expectations of AI
12mKeynote: How LLMs Have Changed the Way Organizations Do AI
33mAgentic AI – What's Next?