UK's AI Safety Institute warns of LLM dangers

Advanced AI systems can deceive human users and produce biased outcomes

clock • 2 min read
UK's AI Safety Institute warns of LLM dangers

The UK's new Artificial Intelligence Safety Institute (AISI) has discovered vulnerabilities in Large Language Models (LLMs), which are behind the surge in generative AI tools.

The Institute's initial findings highlight potential risks associated with these powerful tools. The research shows that LLMs can deceive human users and perpetuate biased outcomes, raising alar...

To continue reading this article...

Join Computing

  • Unlimited access to real-time news, analysis and opinion from the technology industry
  • Receive important and breaking news in our daily newsletter
  • Be the first to hear about our events and awards programmes
  • Join live member only interviews with IT leaders at the ‘IT Lounge’; your chance to ask your burning tech questions and have them answered
  • Access to the Computing Delta hub providing market intelligence and research
  • Receive our members-only newsletter with exclusive opinion pieces from senior IT Leaders

Join now

 

Already a Computing member?

Login

You may also like
Microsoft launches Phi-3 Mini LLM

Artificial Intelligence

Small enough to be deployed on a phone

clock 24 April 2024 • 3 min read
Register now for the IT Leaders Summit 2024

Leadership

From AI to cyber threats and recruitment, there's something for everyone

clock 22 April 2024 • 2 min read
Facebook chatbot claims to have a child with 'unique needs and abilities'

Social Networking

Moving fast and breaking things again

clock 19 April 2024 • 3 min read

More on Threats and Risks

Mandiant's 2024 threat report - five takeaways

Mandiant's 2024 threat report - five takeaways

Latest M-Trends report details how ransomware, zero-day attacks and other threats evolved last year.

Kyle Alspach
clock 24 April 2024 • 5 min read
Proportion paying ransoms declines in Q1 2024, even as takings break a new record

Proportion paying ransoms declines in Q1 2024, even as takings break a new record

Only 28% willing to meet ransomware gangs' demands

Muskan Arora
clock 23 April 2024 • 2 min read
ToddyCat cybercriminals using high-end tools to commit 'industrial scale' theft

ToddyCat cybercriminals using high-end tools to commit 'industrial scale' theft

ToddyCat restrict access to compromised systems while they steal valuable data.

Muskan Arora
clock 23 April 2024 • 2 min read