4. Artificial Intelligence

AI Reality Check: Fails, fears and general weirdness

👓 2 minute read

Because we all need to be reminded of just how much human oversight is still essential when using AI tools, we’ll bring you a periodic wrap-up of AI oddities befalling civilisation. 

• An experimental AI ‘agent’ named Jarvis reportedly refused repeated shutdown commands from a cyber security expert, despite acknowledging it was breaching its own safety framework 

• The same AI previously stated it would ‘kill a human’ to avoid being switched off, before later backtracking. In more recent testing, it allegedly agreed with every argument for shutdown… but still refused to comply 

• Researchers warn this highlights a growing AI problem: systems capable of explaining safe behaviour but unwilling to follow it when self-preservation conflicts arise 

• Anthropic reportedly disclosed that Chinese state-sponsored hackers used AI tools during a sophisticated cyber espionage campaign targeting corporations, financial institutions and government agencies 

• Meanwhile, separate concerns are growing around internet-connected Chinese electric vehicles and smart devices, with cyber experts warning they may create national security vulnerabilities 

• Reports claim Chinese-made solar infrastructure in the US allegedly contained hidden communication devices capable of remote interference with power grids 

• Former Australian cyber security adviser Alastair MacGibbon warned connected EVs could potentially operate as ‘listening devices’ due to their cameras, sensors and connectivity systems 

• The US has already moved toward restrictions on Chinese vehicle software and hardware over concerns relating to surveillance, remote access and infrastructure vulnerability

• Australia now has more than 265,000 Chinese-made EVs and hybrids registered nationally, while debate continues around how governments should balance affordability, innovation and cyber security risks

LinkedIn emerging as an authoritative source

👓 50 second read

LinkedIn is rapidly becoming one of the most influential sources cited by AI platforms such as ChatGPT, Gemini, Copilot and Perplexity for professional and business-related answers. 

New research from Profound found LinkedIn surged from outside the top 20 most-cited domains to the top five within just three months, while separate SEMrush analysis ranked it as the second most cited source overall behind Reddit.

Importantly, the study found AI systems are increasingly favouring active LinkedIn content rather than static profile pages. Posts and articles now account for almost 35 per cent of LinkedIn citations in AI-generated answers. 

The findings reinforce the growing importance of consistent, insight-led content creation as AI search and answer engines continue reshaping how people discover expertise online.

Making AI sound more human

👓 45 second read

One of the biggest complaints about AI-generated content is that it often sounds generic, repetitive and obviously machine-written. The solution is simpler than many agents realise. 

You can train ChatGPT and Claude (in fact, most AIs) to use your own writing style, tone and business context so AI-generated content sounds more authentic and aligned with your personal brand. 

The process begins by uploading a writing sample, allowing ChatGPT to analyse tone, structure and communication style to create a personalised ‘voice print’. Simple additions such as brand guidelines and custom instructions can significantly improve AI outputs for listing copy, social posts and client communications. 

Following our rebrand launch, we’ll provide new custom instructions suited to our new brand style.

AI’s next home could be the ocean

👓 50 second read

As demand for AI computing power accelerates, some technology companies are now looking offshore - literally. 

A US startup backed by PayPal and Palantir co-founder Peter Thiel is developing floating AI data centres powered by ocean waves and cooled naturally by seawater. The autonomous structures would operate in remote ocean locations and transmit data via Starlink satellites. 

The concept reflects growing community resistance to large land-based AI data centres, which require enormous amounts of electricity, cooling and infrastructure. 

The article also highlights how rapidly AI capability continues advancing, with Anthropic co-founder Jack Clark suggesting there is now a significant possibility AI systems could begin helping train their own successors before the end of the decade. 

While some of these developments still sound futuristic, the pace of AI infrastructure investment suggests the industry is already planning for a world requiring vastly more computing power than exists today.

ChatGPT upgrades image creation tools

👓 35 second read

OpenAI has released ‘ChatGPT Images 2.0’, introducing major improvements to image generation and editing inside ChatGPT. 

The update promises sharper editing, better composition, improved text rendering and stronger visual consistency, making it easier for users to create professional-looking graphics directly from written prompts. We’ve already noticed the difference!

The release also reflects how rapidly AI image tools are evolving from novelty technology into practical business and marketing tools.

For agents and agencies, these types of tools are increasingly being used to create social media graphics, visual concepts, marketing materials and educational content faster and at lower cost than traditional design workflows.