Samsung Electronics held its annual AI Forum 2025, bringing together Artificial Intelligence (AI) researchers and industry experts to discuss breakthroughs in artificial intelligence, agentic intelligence, and large language models (LLMs). Day 1 was held at Samsung’s The UniverSE campus, while Day 2 took place online via YouTube live streaming.
In the opening keynote address, Kyungwhoon Cheun, corporate president and CTO of Samsung Electronics’ DX Division, focused the discussion on agentic AI. He explained it in two ways: The first, “advanced reasoning capabilities,” refers to the ability to analyze and solve complex problems step by step, transforming AI from a mere imitator into “an autonomous problem solver.” He also said that Samsung is preparing such AI technologies to deliver value to users.
This article highlights key moments and takeaways from the event, including discussions on why LLMs struggle with reasoning and how researchers are working to prevent illegal AI data scraping and the misuse of generative AI.
‘Sleep-time computing as the next step’
Joseph Gonzalez, co-director of the Sky Computing Lab at the University of California, Berkeley, discussed sleep-time computing, which enables AI systems to prepare using previous context and chat history even before LLMs are queried.
“For me, agents are AI systems that combine reasoning and tool use to accomplish complex tasks,” Gonzalez said. “At the simplest, we can think of it as an LLM interacting with some software services or systems. Increasingly, it’s also interacting with data technologies through things like RAG (Retrieval-Augmented Generation) to find the right information. As we move forward, we start to think about things like context management systems that manage the memory or context exposed to the model.”
He outlined three new efforts related to AI agents and LLMs during his presentation
- The first focuses on sleep-time computation, which allows memory agents to reflect on conversations between interactions, enabling reasoning between test invocations.
- The second involves user agents, which learn human preferences by anticipating what a user will say, effectively building a digital twin.
- The third area covers advisor agents, designed to guide black-box models through complex tasks using reward signals. For context, black-box models are AI/ML systems in which researchers cannot see or easily interpret the internal workings, so they must observe and analyse only the inputs and outputs.
Sleep-time computing involves increasing the computational resources available to a large language model (LLM) during inference (when it’s generating a response) to improve its reasoning and performance on complex tasks. To reduce latency and cost, he said that they are using the idea of background inference, where they “anticipate requests and compute useful quantities even before the query arrives” from the…
Source link
Disclaimer
We strive to uphold the highest ethical standards in all of our reporting and coverage. We blogs.grocliq.com want to be transparent with our readers about any potential conflicts of interest that may arise in our work. It’s possible that some of the investors we feature may have connections to other businesses, including competitors or companies we write about. However, we want to assure our readers that this will not have any impact on the integrity or impartiality of our reporting. We are committed to delivering accurate, unbiased news and information to our audience, and we will continue to uphold our ethics and principles in all of our work. Thank you for your trust and support.
Website Upgradation is going on for any glitch kindly connect at [email protected]
