For this pilot event held at Mama Shelter on Tuesday 29 October, the head of Dev Cloud & AI at Gcore Sergei Grachev, explained that the purpose of this event was to create an IT community in Luxembourg centred around developments in artificial intelligence. It was also a chance for people unfamiliar with tech to try to grasp how it works.

The CEO of LetzAI, , opened the event by quoting a report from Europol: “90% of the content could be AI-generated by 2027.” To have better AI, you need better niche data. What does this mean? If you were to ask a random AI model to create a picture of the Luxembourgish superhero Superjhemp, there’s a high chance it will generate an image of Superman that has a “hemp” logo on it chest. But that doesn’t look like Superjhemp. That’s because most AI models don’t know who Superjhemp is, so they take a wild guess. That’s why it’s important to train AI with niche pictures and context.

Another issue with AI is the complexity of some very long prompts. Let’s say, for instance, we want to see Luc Frieden, wearing a Rolex, surfing, with Kamala Harris riding a dolphin in the bay of San Francisco. The AI model would probably distort the image (such as swap Frieden and Harris’ bodies or recreate a fake bay). Combining many recognisable people in one single picture is one of the biggest challenges right now. LetzAI tries to overlook this problem by giving “intensity” to some parts of the prompt, or giving prominence to one element. 

LeztAI on Wednesday launched a new tool called “Find,” which is a reverse search engine for its AI-generated pictures. This allows people find real products, places or styles from their AI images. Ultimately, it could be used as selling point to businesses whose products are in the picture.

Gcore’s product director AI & cloud Seva Vayner explains about the challenges to come for AI content creators. “There is a need for better infrastructure.” As real-time interactions become more important (such as automated speech recognition, smart cities, virtual avatar or object detection), the processing latency needs to be shortened to between 50ms to 300ms. Servers should be anywhere in the world and closer to the end users. There’s a need to create a global network that offers flexibility and autoscaling according to the demand. That’s what is being done at Gcore with Edge AI, which consists of small servers that can be put anywhere and do smart routing. They establish a connection between the end user and the services through Gcore’s global network backbone--180 points of presence across six continents--allowing a faster response.

Impactful AI cases

For Vayner, one of the most interesting AI cases is the AI Trip Planner, part of a global online travel service. Available in the US, the AI Trip Planner assists travellers in organising trips more interactively. This AI tool aims to offer more tailored and relevant travel options, displayed in a visual format alongside booking details, allowing users to quickly switch between chat-based planning and browsing on the app to finalise reservations.

The Field CTO Belgium & Luxembourg at Dell Technologies, Tom Van Daele, was impressed by Chinese car manufacturers who quickly produce electric vehicles by using digital twins. This allows them to simulate and test designs, identify potential issues before physical manufacturing begins, hence reducing the costs and accelerating the development cycle.

The CTO at Symbiotics, Leminh Nguyen, found the Luxembourg Institute of Health’s voice biomarkers AI to be quite interesting. It aims to use specific features in voice patterns as a means to detect, diagnose and monitor various health conditions.

Strotz is seeing a shift in the business of AI. Generative AI’s rapid evolution is enabling lower-cost content creation across media. Therefore, content creators may want to focus more on protecting their intellectual property and trademarks. That’s why LetzAI has an opt-in approach and reporting features. It’s crucial now with the and concerns such as deepfakes and other ways to misuse content.

Four future trends

According to Nguyen, on-device AI is going to gain popularity. Apple and other tech companies are increasingly integrating AI assistants on their devices, offering faster speeds and offline modes. 

Besides generative AI, Strotz believes augmented reality (AR) glasses might be the next thing in the media hardware segment.

Vayner is betting on autonomous cars and robotics in three to five years. He reckons that ChatGPT5 may be the new Google, the first tool we use when we want to ask a question. 

Van Daele stated that a lot of companies are going on a closed-source model race. He supposes that companies might switch models the same way they switch clouds.