Wednesday, May 29th 2024

Dozen of Experts across Europe, specializing in designing, building and deploying LLMs and related tools. Free of charge.

Agenda Highlights

08:50 - AI Boat + Walk (optional, open to all)

---

---

Change + Training

---

Visiting Unlimitrust residents, 3 AI companies: Effixis, Prometee, +1
Onboarding users and change management

---

Welcome at the Studio

---

Strategy, Design, Innovation + AI, Agile 2.0

---

Introduction from Raphaël Briner, Organiser - Value props of Gen AI

---

Summary made with GPT 4-o
A wide view on the +100 usecases within large organizations supporting a true AI transformation and driving digital AI leadership. Raphaël Briner emphasizes the shift from automation to augmentation with AI, encouraging a broader, enterprise-wide approach rather than limiting AI to specific departments. The speaker highlights the evolution of AI models, particularly the impact of generative AI, and the importance of understanding AI's capabilities beyond simple automation. He stresses the importance of multimodal AI, which combines language and vision models, enhancing agile capabilities and supporting unplanned use cases. He acknowledges AI's application in various domains, from simple to chaotic tasks, and reference research on AI's value propositions in functional, emotional, life-changing, and social impacts.

---

---

Domain-specific innovation

---

Future of Advertising, Media & Legal

---

Summary made with GPT 4-o
Raphaël Briner and Pascal Wicht:
Innovation in AI is likened to a double wave: the massive accumulation of knowledge and the rapid integration into organizations. Comparison with blockchain, noting that AI’s adoption is faster. Future integration across various societal layers, including widespread use in tools like Microsoft’s Copilot and phones. Discussion on job impacts, particularly in translation and customer support, with AI augmenting rather than replacing human labor in most areas.

---

---

Models + Documents + Agents

---

Enterprise Knowledge, Future of search, AI CX

---

Summary made with Claude 3.5 Sonnet in the style of a Tech Mag
Raphaël Briner, Organiser Touched on various AI model types: Foundational or general-purpose models, Vertical and fine-tuned models, Smaller projects using retrieval-augmented generation (RAG). He discussed the typical challenges in the field (AI producing false information, Context loss, The high cost of increasing context windows). Briner introduced concepts like synthetic data, explaining it as AI-generated information useful for various applications. He encouraged attendees to explore RAG systems, suggesting Perplexity as an accessible example.

---

---

Security + Compliance

---

Synthetic Data, Secured LLMs against hackers, Frauds, Risks

---

Summary made with GPT 4-o
Raphaël introduces the complexities of AI models at a gen AI summit. He explains that models are inherently static due to the costly process of creating and updating vectors. Discusses the challenges of integrating new information and the limitations of fine-tuning. Touches on alignment tools, context windows, and special instructions in prompts. Highlights the evolving security landscape in AI, mentioning new players in enterprise stacks and the importance of securing prompts.

---

---

Moderator: Lennig Pedron
While current risks mainly involve reputational damage, future risks could be more severe as LLMs gain autonomy. Opportunities highlighted included improved information analysis, retrieval, and increased efficiency in various tasks. The discussion touched on compliance issues and the risk of AI-generated misinformation. The expanding attack surface of LLMs, both during training and after deployment, was addressed. Speakers debated which actors might lead in AI development, with large tech companies seen as best positioned. The importance of distinguishing between synthetic and non-synthetic content was stressed. The session concluded with hope that democracy would ultimately benefit from AI advancements, citing Taiwan's successful digital democracy implementation as an example.

---

Innovative Models

---

Small models, Natural intelligence & predAI

---

Summary made with Claude 3.5 Sonnet in the style of Wired Mag
Raphaël Briner kicked off the session with another rapid-fire update on the AI landscape, his words tumbling out like an overexcited tech prophet. GPT-4 with voice? Check. Sora and Gemini on the horizon? You bet. Briner's barely contained enthusiasm was infectious as he demoed GPT-4's voice capabilities live, marveling at its speed and latency. But it wasn't all starry-eyed wonder. Briner touched on the darker undercurrents of the AI gold rush - the GPU battles, data collection controversies, and the looming specter of sky-high processing costs. He painted a picture of an AI world where open-source models are nipping at the heels of the closed-source giants, and where specialized models might just be the key to solving thorny problems like data hallucination. Briner's whirlwind tour wrapped up with a nod to the ethical considerations and infrastructure challenges (electricity needs) facing not only the industry but the world. His parting shot? A gleeful endorsement of GROQ.com, as the secret sauce for faster, cheaper AI prompting infrastructure (inferencing). It was a breathless introduction that set the stage for a deep dive into the nitty-gritty of AI innovation.

---

---

The panel, moderated by Johannes David, delved into the challenges faced by smaller AI companies in a field dominated by tech giants. All three panelists emphasized the importance of finding niches and leveraging European strengths like privacy and regulation. The discussion on smaller versus larger models was particularly heated, with Stasenko strongly advocating for the efficiency and practicality of smaller models in enterprise settings. Frey and Dehaye, while not directly competing in the language model space, echoed the sentiment that bigger isn't always better in AI. The conversation took an interesting turn when discussing the future of AI in Europe. Concerns about sovereignty and the risk of decisions being made elsewhere were balanced against opportunities for innovation. Stasenko's warning about the potential disappointment in enterprise AI adoption sparked a debate about managing expectations in the industry. The panel concluded with a brief but thought-provoking discussion on the definition of AGI, revealing diverse perspectives on the future of artificial intelligence. While the panelists differed in their approaches, they shared a common thread of pragmatism and a focus on solving real-world problems with AI, offering a grounded counterpoint to some of the more hyped narratives in the field.

Post-event note from Raphaël: The audience will remember this epic and hearthful message from Anastasia: "It's okay to do boring stuff !".