NVIDIA RTX Spark Makes Its Public Debut at a Korean PC Bang — Joint Demos with KRAFTON, NC, and T1
NVIDIA publicly unveiled the superchip "RTX Spark," announced just recently at COMPUTEX, for the first time in Seoul, South Korea this week. Jensen Huang himself visited T1's PC bang "T1 Base Camp" and presented the RTX Spark alongside Faker (Lee Sang-hyeok), the League of Legends world champion.
RTX Spark is a superchip designed for slim Windows laptops and compact desktop PCs, capable of running AAA titles at 1440p and over 100 fps with ray tracing enabled. It also supports DLSS 4.5 Ray Reconstruction.
At "Optimum Zone PC," a PC bang in Seoul's Gangnam district, NVIDIA ran joint demos with KRAFTON featuring PUBG: BATTLEGROUNDS and Subnautica 2. An AI teammate called "PUBG Ally," built on NVIDIA ACE technology, was also unveiled. At another PC bang, "Portal PC," NC's CINDER CITY and AION 2 were on display, with confirmed support for DLSS 4.5 Dynamic Multi Frame Generation.
Former Naver CEO Han Seong-sook Nominated as South Korean Prime Minister to Lead AI Transformation
South Korean President Lee Jae-myung has nominated Han Seong-sook, former CEO of Naver and current Minister of SMEs and Startups, as Prime Minister. The move is drawing attention as the president's first cabinet appointment.
In a televised press conference on YTN, Presidential Chief of Staff Kang Hun-sik explained that Han was chosen as the person to lead the country's AI transformation and growth, citing her IT industry background and experience at the Ministry of SMEs and Startups. The nomination is widely seen as a signal that South Korea intends to accelerate its national AI strategy.
The Atlantic: Scientific Papers Are "Drowning" in AI-Generated Content
The Atlantic, in an article titled "Science Is Drowning in AI Slop," has exposed the flood of AI-generated content in academic publishing. Reportedly, cases are increasing in which awkward, seemingly AI-generated text and fabricated citations are published in academic journals even after passing the peer review process.
Critics warn that the very credibility of science could be undermined, and recognition is spreading that this is a challenge for the entire academic community.
Developers Confess That "AI Is Rotting Their Brains" — The Reality of Declining Thinking Skills
According to MSN, confessions that "AI is rotting their brains" are pouring in from software developers. The concern is that overreliance on AI coding tools is eroding fundamental programming ability and problem-solving skills.
The topic is also being actively debated on Hacker News, with the developer community facing a dilemma between productivity gains and declining skills.
GraphKV: Compressing the KV Cache by Up to 7.76x with Graph Embeddings
A KV cache optimization project called "GraphKV" has been announced in the LocalLLaMA community. Using an approach that applies graph embedding models, it achieves 7.76x compression on Tiny GPT-2 and 3.36x on Qwen2.5-7B NF4 (32k tokens) while maintaining a cosine similarity of 0.99 or higher.
Inspired by TurboQuant, the project could contribute to improving memory efficiency for local LLMs. Still in its early stages, it is drawing attention as a novel approach to the KV cache problem when handling long contexts.