|
|
|
|
|
|
|
|
|
|
Large Language Models Pose Growing Security Risks
|
|
|
|
|
|
|
|
What's up: Google rolls out a copilot for scientists; Cisco links Salt Tycoon hack to a bug in its software; OpenAI sees big weekly user boost.
|
|
|
|
|
|
Cybersecurity threats are bound to multiply as large language models are commoditized, a process that seemed to take a big leap forward when China’s DeepSeek apparently showed LLMs can be built at lower cost than previously thought. Illustration: Thomas R. Lechleiter/WSJ
|
|
|
|
|
|
Good morning, CIOs. Cybersecurity risks posed by large language models are bound to multiply with their commoditization as an ever-growing number of LLMs find their way into corporate systems.
Those risks include bringing unsafe code or data into a company as well as exposing sensitive corporate or personal information to models widely accessible to the public.
And in the absence of government leadership on the matter, companies, by and large, should consider themselves on their own.
So what’s a technology leader to do? WSJ Columnist Steven Rosenbush talked to a number of tech leaders on the processes needed for securing LLMs, starting and ending with human oversight.
|
|
|
|
|
The most important thing for companies to understand is where their responsibility lies at any given moment or in any given situation, and to verify their tech suppliers and partners are holding up their end of the bargain.
|
|
|
|
|
|
|
|
|
|
Content from our sponsor: Deloitte
|
|
|
Data Architecture: 6 Ways to Navigate a Shifting Landscape
|
|
To help data systems remain resilient, efficient, and future-ready, technology and business leaders can consider six key actions. Read More
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Lenovo sees DeepSeek breakthrough as fuel for PC sales. The world’s largest PC maker said that it has integrated the Chinese startup’s latest technology into its devices. Chief Executive Yang Yuanqing tells the WSJ that AI PCs will make up 25% of Lenovo’s global PC shipments in 2025.
|
|
|
OpenAI claims 400 million weekly active users as of February. That’s up 33% from 300 million in December, according to Chief Operating Officer Brad Lightcap. (CNBC)
|
|
|
It’s “game on” for AI. Microsoft in Nature said its World and Human Action Model, trained on human gameplay data and designed to generate 3-D sequences in response to a player's actions, points to AI's potential to play an ever larger role in creative ideation in other domains, including music and video.
|
|
|
Together AI valued at $3.3 billion after $305 million funding round. The company’s platform offers developers access to open-source AI models and the necessary computing power to build applications. (Bloomberg)
|
|
|
|
|
|
|
|
|
|
Illustration: Allan Sanders
|
|
|
|
|
AI Buzzword of the Day: 'Vibes'
|
|
|
|
|
Tech industry insiders are increasingly relying on intuition, rather than hard data, to judge which AI chatbots are best.
Are a chatbot’s responses concise or overwrought? Pandering or confrontational? Researchers call this a “vibe check,” and to many, it’s just as important as academic benchmarks that measure the ability of AI to answer multiple-choice questions or perform complex tasks.
“Vibe checks are great evals,” OpenAI President Greg Brockman recently wrote on X, using the industry shorthand for evaluations.
|
|
|
|
|
|
|
|
|
