The Gartner IT Symposium shows it’s time to trust specialized AI
The Gartner IT Symposium is a worldwide event for global IT leaders. Attendees flock to its venues around the world to tap into the latest thinking on how technology strategy intersects with business strategy. When DeepL presented at Gartner symposia in Orlando, Barcelona and Tokyo last month, one theme continued to come up in discussions: When it comes to AI, if you want real business value, you need to call in the specialists.
Generic AI models like ChatGPT-4 and Google Gemini, which many people use in daily life, don’t deliver on what matters most for enterprise organizations. Those models are trained on vast reams of publicly available data, and they’re developed for a broad set of unrelated functions. They’re jacks of all trades that can do a range of things fairly well. However, when your business depends on accuracy, consistency and security, “fairly well” won’t cut it.
The regular hallucinations that generic AI models create can be confusing and irritating for an individual experimenting with them. For an enterprise organization, they can be disastrous. Mistakes throw off experiences for thousands or millions of customers, undermine brands and set strategies off in the wrong direction. For these and many other reasons, business leaders and their employees can’t be fully confident about generic AI outputs. And when they have to keep second-guessing their AI outputs, it’s impossible for those platforms to deliver real value.
The difference is in the data — and the training
The big difference with a specialized AI platform is that it’s built, developed and trained for a specific purpose, and uses carefully curated, domain-specific data. You don’t ask a carpenter to do heart surgery just because they’re good at cutting things precisely. Similarly, when you’re developing an AI for medical imaging and diagnosis, you don’t train it on every type of image you find online and then trust it to be good at spotting tumors. Instead, you train it on the specific, relevant medical images and other data that it needs.
Specialized AI models require investment in bespoke model development and training. That investment goes on to generate major returns through the deep understanding the models have, and the range of relevant use cases they can expedite. It generates returns, too, in terms of cost and energy efficiency.
Why language is the ultimate use case for specialized AI
As DeepL’s CEO Jarek Kutylowski explained recently on the Unsupervised Learning podcast, one of the priorities for IT leaders is identifying which AI use cases can deliver financial returns that justify investment in specialized models. It’s abundantly clear that language is one of those use cases. Deeper understanding of language enables a wide range of use cases; customization to fit the terminology of each business; and a broader range of innovative solutions. When it comes to language, the difference in quality between generic and specialized models is undeniable — and so are the returns that come from greater accuracy.
For Harvard Business Publishing (HBP), those returns come from the ability to scale localization and deliver an archive of content to global audiences in their preferred language. At DeepL Dialogues, HPB’s Associate Director Multilingual Services, Derick Fajardo, explained how DeepL had rapidly increased the pace of localization, automatically translating over 25,000 articles with a quality that readers appreciate and the business can trust.
For Brioche Pasquier, an early adopter of DeepL Voice, real-time subtitles in every participant’s preferred language are transforming the experience of multilingual, virtual meetings. International Coordinator Christine Aubry explained Brioche Pasquier’s commitment that nobody should be excluded from projects and tasks because of language barriers. Enabling everybody to follow and participate in meetings makes that possible.
Specialized understanding drives innovation
Specialized LLMs like DeepL’s don’t just deliver greater accuracy. They also enable wider innovation, bringing benefits to more areas of a business. Features like DeepL Voice, DeepL Write, and automatically generated glossaries are made possible by an understanding of the context, rhythms, signals and structures of different languages — an understanding that goes beyond just spotting patterns through repetition. This deeper understanding of language enables DeepL to anticipate what’s being said, imagine different ways to say it within a particular cultural context, and tailor translations to fit a specific brand and business.
In the years since DeepL first launched, the value of such specialized Language AI has increased exponentially. Increasing international trade creates more diverse and multilingual companies, workforces, and customer engagements. Every touchpoint between people with different preferred languages creates opportunities for enterprises to deliver value through specialized AI they can trust. Those are the forms of AI that IT leaders are increasingly turning to.