Unbabel CEO's Insights on AI Trends & Entrepreneurship!
View original at youtube.comVia News YouTube - Video Interview Title: Unbabel CEO's Insights on AI Trends & Entrepreneurship!…
What we drew from this source
The claims Via News extracted from this document. We point to the source; we don't replace it.
AI-first companies built on top of ChatGPT risk being made obsolete when OpenAI launches a similar feature; companies with existing customers benefit most from deploying AI.
60% confidencePedro is a product CEO rather than a sales CEO and needs a strong sales counterpart to cover his blind spots.
60% confidenceUnbabel's biggest early challenge was naivete: they underestimated the complexity of enterprise translation and of recruiting, onboarding, curating and evaluating human translators.
60% confidenceScaling is about repetition: find product-market fit, prove a thesis, then do the same thing repeatedly and better; embrace what is not sexy.
60% confidencePedro works out six days a week, reserves mornings for deep work and afternoons for quick meetings, and recommends a chief of staff or great EA.
60% confidenceMiguel stated that Unbabel has reached 49 million in revenue and has half a million customers.
60% confidenceHumans will evolve from playing one instrument to conducting multiple AIs, and human in the loop as data generation will continue for a long time.
60% confidenceUnbabel anonymizes PII in every incoming text, replacing names with same-gender names to avoid translation errors, and holds ISO certifications.
60% confidenceHalo has reached about 15 words per minute, versus about 2 words per minute for Stephen Hawking.
60% confidenceQuality estimation lets Unbabel build translation pipelines from fully automatic to fully human, depending on use case, content type and language pair.
60% confidenceAround 2026, translation may merge with content creation if AI is trusted to generate content in multiple languages from the start, but this is not close; AI is a cognitive prosthetic.
60% confidenceIn 2024 AI adoption will lower revenue per word but increase words translated, so the translation market will expand.
60% confidenceUnbabel used ChatGPT in production, then built its own fine-tuned LLM-based model, and is now training its own model from scratch, favouring smaller task-specific models over giant general-purpose ones.
60% confidence
