| Mobile Web

Search results for University of California Berkeley

AI & Enterprise

Cheap AI does not always save money; total cost can be higher, study finds

A study found low-priced AI models are not always cost-efficient. After running more than 6,800 tasks across eight advanced AI models, researchers found total costs for cheaper models were higher in 32 percent of cases. They said token prices and per-call fees can be misleading because models may require very different amounts of reasoning and action steps. Repeated runs of the same model also showed wide cost swings.

Industry

Ryu Jae-cheol (류재철) says LG Electronics making physical AI bet to drive meaningful change

Ryu Jae-cheol, CEO of LG Electronics, hosted a North America Tech Conference in San Francisco on Aug. 29 to secure overseas R&D talent, the company said. About 100 participants attended, including PhD researchers from major U.S. universities and experienced R&D staff from big tech and startups. LG Electronics presented robotics, AI data center cooling and mobility as key growth pillars based on physical AI, and outlined data and manufacturing strengths.

Industry

Experts warn flashy humanoid robot stunt videos should be viewed with suspicion

Flashy humanoid robot demonstration videos are spreading online, but they should not be used to judge real work capability, experts say. Industry and academic figures cited by Ars Technica point to a gap between viral clips and on-site performance. They stress repeatability across environments and broad usefulness, not one-off stunts. They also urge scrutiny of autonomy, demo settings and playback speed, and call for large-scale quantitative validation in real conditions.