Samsung just opened its ninth annual AI Forum 2025, bringing together the world's leading AI researchers for a two-day deep dive into the technology reshaping everything from chip design to consumer devices. The event features keynotes from deep learning pioneer Yoshua Bengio and Stanford's Stefano Ermon, signaling Samsung's push beyond generative AI into autonomous agent systems that can make decisions without human oversight.
Samsung Electronics is making its biggest AI research play yet. The company's AI Forum 2025, which kicked off today in Yongin, Korea, reads like a who's who of artificial intelligence - from Turing Award winner Yoshua Bengio to Stanford's diffusion model pioneer Stefano Ermon. But this isn't just academic theater. Samsung's betting the future on what it calls 'agentic AI' - systems smart enough to work autonomously.
"Samsung is applying AI across our operations to develop foundational technologies that make AI more intuitive and seamless," Samsung Vice Chairman and CEO Young Hyun Jun told the opening session, according to company statements. The forum's timing isn't coincidental - it comes as Samsung faces mounting pressure from Apple and Google in the AI-powered device race.
The first day dove straight into semiconductor AI applications, with Professor Bengio delivering what attendees described as a sobering keynote about current AI risks. His solution? Scientist AI, a new model designed to prioritize factual accuracy over human-pleasing responses. "Unlike models built to mimic or please humans, Scientist AI focuses on providing truthful answers grounded in verified facts and data," Bengio explained, addressing growing concerns about AI hallucination and control problems.
Samsung's semiconductor division showcased how AI is already transforming chip design workflows. Yong Ho Song, who heads the company's DS Division AI Center, put it bluntly: "AI is already an essential tool in chip design and software development. As semiconductor manufacturing grows more complex, we expect AI to help address the technical challenges that arise." The company's Yongin facility, where day one took place, is ground zero for Samsung's AI chip ambitions as it competes with Nvidia in the inference market.
But day two is where Samsung really shows its cards. The focus shifts to "Generative to Agentic AI" - systems that don't just respond to prompts but actively make decisions and carry out tasks. Paul Kyungwhoon Cheun, Samsung's DX Division CTO, framed it as the next evolutionary leap: "As we enter the era of Agentic AI, Samsung will continue to focus on developing AI technologies that provide users with tangible benefits."
The technical sessions reveal Samsung's strategy across its product ecosystem. The company demonstrated on-device AI technologies designed to bring large language models directly to smartphones and TVs, automatic dubbing that preserves original speaker voices, and multi-agent systems for document processing. One standout: an on-device AI studio for developers that could dramatically speed up generative AI development cycles.
UC Berkeley's Joseph Gonzalez will present research on "sleep-time compute" - letting AI agents use downtime between interactions to reason and plan. Arizona State's Subbarao Kambhampati tackles what he calls "large reasoning models" to fix current AI limitations in factual accuracy and complex reasoning. These aren't just academic exercises - they're roadmaps for Samsung's product development.
The forum also recognized this year's Samsung AI Researcher awards, going to University of Toronto's Nicolas Papernot, UC San Diego's Rose Yu, and NYU's Lerrel Pinto. But the real story is how Samsung is positioning itself not just as a hardware company that happens to use AI, but as an AI company that happens to make hardware.
What's striking is Samsung's timing. While OpenAI dominates headlines and Google pushes Gemini integration, Samsung is quietly building the infrastructure for AI that works offline, on your device, without cloud dependencies. The camera AI for automatic color temperature adjustment and knowledge distillation for more efficient LLM training aren't flashy, but they're exactly the kind of practical AI applications that could differentiate Samsung devices in a crowded market.
The event streams live on Samsung Developer's YouTube channel, making day two accessible globally. That's significant - Samsung is broadcasting its AI capabilities to developers worldwide, essentially saying 'here's what we're building, come work with us.'
Samsung's forum arrives as the AI industry faces growing questions about sustainability, safety, and practical applications beyond chatbots. The company's focus on on-device processing and specialized applications like automatic dubbing suggests it sees opportunity in solving real-world problems rather than chasing the latest language model benchmarks.
Samsung's AI Forum 2025 signals the company's evolution from hardware manufacturer to AI platform builder. By bringing together top researchers and showcasing practical applications from chip design to consumer devices, Samsung is positioning itself for the next phase of AI adoption - where intelligence runs locally, works automatically, and integrates seamlessly into daily life. The question isn't whether agentic AI will reshape our relationship with technology, but whether Samsung can execute on this vision faster than its Silicon Valley competitors.
