How is AI changing the way we transform and generate content? This was the question tackled by the SlatorCon panel moderated by Slator’s Florian Faes, which brought together Michel Lopez, Founder and CEO of e2f, Konstantin Savenkov, CEO of Intento, and Jochen Hummel, CEO of Coreon and ESTeam.
The panelists exchanged broad views on the multibillion-dollar digital content creation market. The discussion kicked off with translation memories (TMs) as AI training data for a wide range of use cases, from machine translation to named entity recognition (NER).
AI’s role in transforming content — from speech into text or vice versa — was then examined, with Savenkov and Lopez arguing for a new perspective: we can see these activities as composed of successive transformation steps, each with the potential to be enhanced by AI.
In subtitling, for example, one can use automatic speech recognition (ASR) to transform voice into text, switch the language using machine translation (MT), and then use an AI tool to cut the text into pieces. Plus, one has the option of applying AI-powered QA tools at any point along the way to improve the quality.
Era of AI-Generated Content Is Upon Us
The panel then took on the topic of content generation and the world’s largest language model, GPT-3. The speakers drew parallels between MT and GPT-3 and touched on how GPT-3 has already been put to practical use.
Hummel tipped the audience off to a fascinating short movie for which the script was written by GPT-3.
Below are a few select highlights from the session.
Slator: How big a deal is GPT-3?
Savenkov: It feels like when Google came out with neural machine translation in 2016 — but now for content generation. If you’re someone who feels like you missed the MT wave, and think you’d do something differently next time, well, this is your time.
Hummel: What GPT-3 does is amazing, but it also does some funny stuff. You can’t just take the content and publish it — you need a human in the loop. The parallels between the localization industry and the content creation industry are amazing.
Slator: From a practical point of view, are you already using GPT-3?
Hummel: Of course, it’s already been put into practice. I’ve recently made an angel investment in a company that is automating the writing of social media posts.
Lopez: We’re actually generating more content than we are translating. But it’s not for human consumption. We’ve generated about three million words of text for scripts, for the purpose of training speech recognition engines.
Watch the “Content Creation in the Age of AI” panel discussion and the full SlatorCon Remote September 2021 event on demand, here.