Researchers identify fundamental AI security flaws, a geothermal plant is revived in New Mexico, and SK Hynix employees receive significant profit-sharing bonuses.

A team of researchers has identified a fundamental security flaw in large language models that may be impossible to fully resolve. Meanwhile, in the climate tech sector, the Zanskar company has successfully restored a failing geothermal power plant in New Mexico to full capacity. Additionally, semiconductor workers at SK Hynix in South Korea are seeing a surge in personal wealth following record-breaking profits from the AI chip boom.

Fundamental Security Flaws in LLMs

Researchers presenting at a recent AI conference argued that large language models cannot be made fully secure due to a core issue in how they process instructions. By exploiting this flaw, the researchers demonstrated that models could be manipulated into providing restricted information, such as instructions for synthesizing cocaine or sabotaging aircraft navigation systems.

Reviving Geothermal Power

In June 2024, Zanskar acquired a struggling geothermal plant in New Mexico that was becoming uneconomical due to cooling water reservoirs. Two years later, the company has returned the facility to full capacity by utilizing advanced modeling and modern drilling techniques to identify and tap a new, more productive well site.

Semiconductor Industry Bonuses

Employees at the South Korean semiconductor firm SK Hynix have received substantial bonuses following high profits driven by the AI chip market. The company agreed to distribute 10% of its operating profits to staff, resulting in payouts of approximately $476,000 per employee this year.