How I Tried to Reinvent the Wheel And Got Crushed by a Hyperscaler

Lessons From a True Story

This post is a little different: it’s a self-reflection on my first big project—building an industrial machine-translation system. You can find a description of that system in this paper. On the surface everything looked great, but in reality… oh, what a mess. The experience taught me a lot about the buy-vs-build dilemma, about designing for operations, and about not getting carried away by pure research. The system itself is long gone, yet I keep seeing people repeat the same mistakes with today’s generative-AI wave. After all, machine translation is where neural language modelling began, and what we see now with ChatGPT, MT specialists lived through years before the hype. This is an embarrassing and educational tale for all the industrial AI specialists.