Language, the often-overlooked heart of a digital customer experience, offers brands the opportunity to scale their products and services for the global market. Utilizing AI for localization allows businesses to generate high-quality copy with greater efficiency—an asset for companies of every size and industry. 

In our recent AI Day Webinar, our panel of AI and localization experts sheds light on the mystery surrounding LLM and Contextual AI. Lilt's 90-minute AI sessions bridged the gap between cutting-edge AI technology, experienced linguists, and your business needs, and showcased a few speakers who have revolutionized the industry.

Whether you watch the 90-minute session or explore this summary, you will walk away with a concrete foundation in AI translation to help you transform the way you execute localization.

Session 1: Excitement around GPT4 + AI Translations 

The first session featured speakers Andy Jolls (CMO of Lilt), Spence Green (Lilt’s co-founder and CEO), and Sarah Sandberg (Lilt Senior Product Manager). Green kicked things off with some information on the humble beginnings of Lilt’s AI for Localization, stating: 

“—Lilt came from the Stanford AI lab many years ago. We had two fundamental goals. One is that we wanted to get the most advanced AI technology into the hands of business users so that they could derive value from it.” 

Green went on to highlight how Lilt’s early arrival to market for these systems has allowed Lilt to develop AI that provides businesses with an immediate return on investment.

Sandberg provided a live demonstration of Lilt’s Contextual AI Engine, showcasing the three key product features: 

• Adaptation  
• Prediction 
• Comparison. 

Green went on to describe how GPT-4 will alter existing localization workflows, explaining:

“I think that people are going to be really excited by how much more fluent the system's become, and it's gonna be the same level of improvement that we saw in 2016 and 2017 when we shifted to using neural network approaches. I also think that what's really exciting is these systems can be used for a lot of other interesting tasks in enterprise language, like quality assurance and text generation from scratch…we're working on some products to that extent that are going to really change fundamentally the workflows—the create, translate, publish workflow that has existed for many, many years.”

During the Q&A, Green offered an essential takeaway when asked how Lilt’s AI system compares to Chat GPT, stating: 

“I want to really encourage people to compare the effectiveness of In-Context Learning between various systems, and to not get so focused on differences between different providers because when you operationalize this in practice, you need In-Context Learning to drive maximum value.”

Green also defined In-Context Learning,

“...the basic idea of in-context learning is that you learn in real-time on data…what you want to do is have a system that learns from [every demonstration so that]  its next prediction is better.”

With In-Context Learning, your AI for localization strategy has the advantage of a system that learns from linguist feedback as it goes.

Session 2: Demystifying Large Language Models + the Impact of using Connectors to Accelerate your Business 

For Session Two, Brittney Benchoff (Lilt Head of Product) was joined by Joern Wuebker (Lilt Director of Research) and Tim Gallant (Lilt Senior Staff Engineer).

Wuebker kicked off Session Two by explaining how Contextual AI works and why it's essential to training LLMs: 

“It's a class of artificial intelligence systems that can understand and interpret the context of basically any given situation or query, and adjust its responses to be appropriate to the situation—right, to be contextualized. And when we look at our human linguist workflow, which is visualized here, you can see at the heart we have Lilt’s Contextual AI Engine, which is basically a large language model under the hood, similar to GPT-4.”

Wuebker went on to provide some facts on Lilt’s contextual AI engine, including:​​

​​• It is under the hood of a so-called encoder decoder transformer architecture, which is similar to GPT-4.
• Lilt’s model is more than 1000x more compact than GPT-4, which means it can be deployed for real-time in-context learning on real-time prompt completion with low latency in real-time by linguists and can be deployed on a customer’s own computing infrastructure.
• In-context learning is achieved by adapter layers, a technology introduced or invented by Google researchers in 2019, which eliminates the need for manual retraining cycles.

Wuebker provided a Transformer Model and explained the visualization’s significance, saying:

“The end coder's job [on the left-hand side] is to take a sentence in the source language and basically transform it into a list of numbers. The numbers are how the model interprets and understands the meaning of a sentence…The decoder is on the right-hand side, is to take these numbers, interpret their meaning as the meaning of a sentence, and transform them into a readable, human language sentence in the target language.”

Wuebker provided statistics on how Lilt’s AI for Localization measures up, stating:

“..with the word prediction accuracy of 85%...in terms of finances, that basically means you can, translate the same or we can translate twice the amount of data, twice the volume with the same budget, or allowing even larger volumes of content to be translated even faster.”

Gallant provided another demonstration of the Lilt platform benefits clients including Figma and Shopify.

Session 3: The Future of Work with AI Technology

Session Three welcomed Andreas Laursen (Lilt Director of Program Management) and Jesse Rosenbaum (Lilt Security and Compliance Specialist). Laursen addresses some of the concerns and anxieties that may come with new technology, as well as the fear that quality may be sacrificed with the efficiency of AI technology.

Laursen offers solace on what contextual AI means for linguists, stating:

“With linguists, AI represents a technology they can actually learn and invest in and really specialize in and become very quick. And that gives them control to really set their own terms. They can become very good at AI and very productive, and that delivers massive value to customers. And that makes them, you know, even more in demand.”

Rosenbaum comments on the high standard of safety and security Lilt’s AI system possesses, saying:

“The government has some of the highest standards imaginable for security and compliance. And so if we're able to deploy on these really sensitive and secure systems, it meets the commercial requirements by a long shot.”

During the Q&A, CEO Green further supported Lilt’s overall security by adding,

“We build and own all of the software that's used, including the way that we distribute data to our translation supply chain. So our enterprise customers can be assured that they have complete control over your data and how it's used.”

Revolutionize your AI for Localization with Lilt

With many AI technologies out there, companies must do their due diligence and choose the right AI solution. No matter what role you’re in, whether you’re managing programs or you’re a translator, we hope you learned about the future of AI and language. 

A lot of people have been waiting to make the jump into AI tools, and now is the best time to do it. Lilt is the partner to help you with AI translation for all your content and media needs. We've got the most experience in deploying AI systems to bring the latest state-of-the-art technology into enterprise language. 

Whether you’re considering implementing an AI approach in the enterprise or have already started the process, this webinar nicely summarizes all the topics you should consider when selecting your partner.

Let Lilt make elevating your AI for Localization strategy a frictionless transition. Request a demo today and see Lilt’s AI platform in action. We help you make the business case to your leadership team and have a structured change management program to help you with all the internal changes to your systems and processes.