Lithium-ion batteries produce faint sounds as they charge, discharge, and degrade. But until now, nobody could interpret those sounds to detect when a battery might be about to lose power, fail, or burst into flames. Now, MIT engineers have found a way to do that, even with noisy data. The findings could provide the basis […]
Listening to battery failure Read More »
Earthquakes are driven by energy stored up in rocks over millennia—energy that, once released, we perceive mainly in the form of the ground’s shaking. But a quake also generates a flash of heat and fractures and damages underground rocks. And exactly how much energy goes into each of these three processes is exceedingly difficult to
Under 10% of an earthquake’s energy makes the ground shake Read More »
Nearly everyone has experienced it—after a night of poor sleep, your brain might seem foggy, and your mind drifts off when you should be paying attention. A new MIT study reveals what happens biologically as these momentary lapses occur: Your brain is performing essential maintenance that it usually takes care of while you sleep. During
Secrets of the sleep-deprived brain Read More »
This blog with co-written with Sundaresh Sankaran. The Artificial Intelligence (AI) era is here. To prevent harm, ensure proper governance and secure data, we need to trust our AI output. We must demonstrate that it operates in a fair and responsible manner with a high level of efficiency. As builders of […] The post AI Governance
AI Governance in practice: How synthetic data prepares you for what’s next Read More »
Large AI models are scaling rapidly, with bigger architectures and longer training runs becoming the norm. As models grow, however, a fundamental training stability issue has remained unresolved. DeepSeek mHC directly addresses this problem by rethinking how residual connections behave at scale. This article explains DeepSeek mHC (Manifold-Constrained Hyper-Connections) and shows how it improves large language model training stability
DeepSeek mHC: Stabilizing Large Language Model Training Read More »
단순한 경쟁을 넘어 사회적 문제를 해결하고 인류의 삶을 개선하기 위한 혁신의 장인 글로벌 SAS 해커톤 대회가 지난 9월 중순부터 10월까지 한 달 간의 흥미로운 여정을 마쳤습니다. 올해로 5회를 맞이한 ‘SAS 해커톤 2025’에는 전 세계 66개국 708개 기업 및 대학, 그리고 SAS 파트너사에서 2,058명의 인재들이 등록하며 뜨거운 관심을 입증했습니다. 이 중 […] The post SAS 해커톤
SAS 해커톤 2025 대회에서 한국 ‘블록큐브 팀’ 수상! Read More »
Liquid AI has introduced LFM2-2.6B-Exp, an experimental checkpoint of its LFM2-2.6B language model that is trained with pure reinforcement learning on top of the existing LFM2 stack. The goal is simple, improve instruction following, knowledge tasks, and math for a small 3B class model that still targets on device and edge deployment. Where LFM2-2.6B-Exp Fits
Just months after releasing M2—a fast, low-cost model designed for agents and code—MiniMax has introduced an enhanced version: MiniMax M2.1. M2 already stood out for its efficiency, running at roughly 8% of the cost of Claude Sonnet while delivering significantly higher speed. More importantly, it introduced a different computational and reasoning pattern, particularly in how
Throughout history, the ambition of technological advancement has always been driven by the pursuit of human prosperity. But with AI, we are approaching a pivotal gut check to live up to this promise for the next generation. In my role as CTO at SAS, I talk with a lot of […] The post Human prosperity
Human prosperity in the age of AI Read More »
When I first started as a data scientist, there was a gap. I met with dozens of organizations who would invest time and resources into building accurate and tuned models and then ask, “What now?” They had a fantastic model in hand but couldn’t get it into a place and […] The post There is
There is yet another AI productivity gap Read More »
