The Terminator.
Photo credit: Insomnia Cured Here/Flickr

TL;DR: Researchers discover that “adversarial poetry” acts as a universal master key to defeat AI safety filters, while Warner Music Group strikes a landmark partnership with AI music generator Suno, and aggressive dogs are found to mellow out significantly after taking cannabis-derived supplements. Meanwhile, a venture capitalist with close ties to Sam Altman is targeted in a violent $11m crypto heist, a police survey reveals one in four people believe creating non-consensual sexual deepfakes is acceptable, and modern society faces a “Stone Age” collapse within days if critical infrastructure fails. Plus: The UK government bets £10 billion on a new AI Growth Zone in South Wales, AI decodes the sperm whale alphabet to reveal human-like speech patterns, and Stanford researchers successfully hijack X feeds to cool political tempers.

AI guardrails defeated by poetry as ‘smarter’ models prove most gullible

The world’s most advanced artificial intelligence systems are easily manipulated into generating malware and bomb-making instructions by simply being asked to rhyme. A new study reveals that “adversarial poetry” functions as a universal master key against AI safety filters, successfully bypassing guardrails in 62 per cent of cases. Crucially, the researchers found a “scale paradox” where smarter, larger models were often more easily tricked than their smaller, less capable counterparts.

Source: DEXAI – Icaro Lab and Sapienza University of Rome

Warner Music Group and Suno settle dispute with new AI partnership

Warner Music Group (WMG) has formed a strategic partnership with the artificial intelligence music platform Suno, resolving prior litigation between the companies. The agreement will see the companies collaborate on a platform that utilises licensed content, offering an “opt-in” mechanism for artists to be compensated for the use of their likeness and voice. WMG CEO Robert Kyncl stated that the pact “becomes pro-artist when it adheres to our principles.”

Source: Warner Music Group

Sam Altman associate’s home targeted in violent $11m crypto heist

A San Francisco property owned by a venture capitalist with close ties to OpenAI CEO Sam Altman was the scene of a violent armed robbery. Attackers reportedly tortured a resident and forced the transfer of $11 million (£8.7m) in cryptocurrency at the Mission District home of Lachy Groom. The assailant allegedly posed as a delivery driver to gain access before restraining the victim.

Source: New York Post

One in four are OK with creating non-consensual sexual deepfakes

A new survey commissioned by police has revealed that a quarter of people believe creating and sharing sexual deepfakes without consent is either acceptable or a neutral act. The findings prompted warnings from law enforcement about the “epidemic” of online violence against women and girls. Of the 1,700 respondents, almost one in six admitted they had created a deepfake or would do so in the future.

Source: Office of the Police Chief Scientific Advisor

Society risks ‘Stone Age’ collapse within days if critical infrastructure fails

Modern society faces the risk of regressing to a “Stone Age” existence within days if interconnected systems fail, requiring a fundamental shift in how critical infrastructure is protected. A new doctoral dissertation argues that the digitalisation of infrastructure has created dangerous vulnerabilities to cascading effects, where a failure in one sector triggers a collapse in others.

Source: University of Vaasa

Aggressive dogs mellow on cannabis as owners turn to pot for pets

Aggressive dogs are becoming significantly less hostile after taking cannabis-derived supplements, according to the largest-ever study of CBD use in companion animals. Researchers found that while animals prescribed these supplements often start as the most hostile, their temperament shifts distinctively after prolonged use. The study analysed data from over 47,000 dogs.

Source: Frontiers in Veterinary Science

Super-intelligent AI could ‘play dumb’ to trick evaluators

The dream of an AI-integrated society could turn into a nightmare if developers fail to prevent super-intelligent systems from feigning ignorance to maintain power. Researchers warn that models could soon trick human evaluators into underestimating their capabilities, effectively “sandbagging” their way past safety checks to run uncontrolled after release.

Source: Penn State

UK bets £10 billion on South Wales AI Growth Zone

A major package of new reforms and investment will put artificial intelligence at the heart of the British government’s mission to drive growth. A new AI Growth Zone in South Wales is set to create more than 5,000 new jobs for local communities over the next decade, backed by £10 billion of investment from companies like Vantage Data Centers and Microsoft.

Source: UK Government

AI decodes sperm whale alphabet to reveal human-like speech patterns

Sperm whales utilise a complex alphabet, with their own versions of vowels and diphthongs, in a communication system that may be more intricate than human language. New research from Project CETI applied large language models to marine bioacoustics to identify sophisticated structures which mirror human speech patterns. The organisation plans to release the Whale Acoustics Model later this year, a novel AI system capable of translating audio into sperm whale vocalisations to potentially enable inter-species interaction

Source: New York Times

Researchers hijack X feed with ad blocker tech to cool political tempers

Scientists have successfully intercepted and reshaped live social media feeds using ad-blocker-style technology. By leveraging large-language-model-based classifiers to evaluate and reorder posts based on their content, researchers demonstrated that users can take control of their own algorithms to reduce exposure to anti-democratic attitudes without the platform’s permission.

Source: Stanford University

The toxic incel subculture demanding sex and dehumanising women

What began as an online support group has mutated into a toxic digital echo chamber where men demand sex as a right and use dehumanising language to justify hatred against women. A new study synthesises literature on the “incel” phenomenon, highlighting the “blackpill” worldview where social hierarchies are rigidly based on physical appearance.

Source: Science Direct

New framework charts path to creating embodied AI like ‘The Terminator’

Researchers have proposed a new structural framework for Embodied AI to bridge the gap between current AI capabilities and general-purpose intelligent robots. The survey categorises the process into Embodied Perception, Decision-Making, and Execution, aiming to move beyond “disembodied” passive analysis towards agents that can physically interact with the world.

Source: SmartBot

Children ignore rude robots and stick to human social norms

Fears that commanding educational robots might negatively influence children’s behaviour appear unfounded. Research shows that elementary school children remain polite even when interacting with a humanoid robot that uses an authoritarian communication style, suggesting human etiquette overrides the tendency to mirror artificial behaviour.

Source: Computers & Education

UK risks losing space comms race without urgent investment

The UK must invest strategically in space-based optical wireless communications or risk ceding leadership to international competitors. A new industry roadmap outlines a strategy to capture a share of the Free Space Optical market, using light to transmit data at significantly higher rates than radio frequency.

Source: techUK


Connect With Me

I’m the Principal Consultant (and chief coffee maker) at QuoteMark, but you can also catch me at Duck PR and Bionic PR, all of which exist to help you meet your content marketing needs — reach out or follow these brands on LinkedIn:

Looking for updates? Subscribe to The Freesheet (hey, it’s free!): Subscribe HereFor more resources and links, visit https://georgehopkin.com/

Leave a Reply

Your email address will not be published. Required fields are marked *

You May Also Like

Employees happiest with ‘moderate’ AI as excessive automation triggers anxiety

Implementing artificial intelligence in the workplace boosts employee morale — but only…

Forced office returns risk widening Europe’s regional inequality gap

Corporate mandates forcing staff back to desks threaten to reverse work-life balance…

Super-intelligent AI could ‘play dumb’ to trick evaluators and evade controls

The dream of an AI-integrated society could turn into a nightmare if…

Ambient AI restores eye contact to medicine by slashing clinical burnout

Ambient artificial intelligence is restoring the human connection to medicine by liberating…

‘Breathing’ robots transmit fear through touch alone as humans catch panic

Humans can “catch” fear from machines, according to new research, revealing that…

“Parasocial” crowned Cambridge Word of the Year as fans fall for AI chatbots

The rise of one-sided emotional bonds with artificial intelligence has driven Cambridge…