Bad AI: Marathon Bots
Apr 23, 2025AI IS ACCELERATING - GET READY
Let's help you understand the WTF of AI and how it helps YOU and your organisation grow. Ready, set, ethics.
This week:
π± Bad AI: Marathon Bots
π Event Tomorrow: Unethical AI
π₯ The Future: AI Morality
β¨ AI Advances: ChatGPT Model Updates
π‘ Trends & Insights
BAD AI: Marathon Bots
 |
It's started. Twenty humanoid bots joined hundreds of human runners at the Yizhuang half-marathon in Beijing.
β
So what happened? Are we fitness-cooked then?
β
Not quite. Some bots overheated and collapsed mid-race - tbf, a relatable human burnout arc - while others just power-walked menacingly toward the finish like they were late for a silicon coup. At least they didn't win. This time.
β
Officials called it a landmark for human-robot harmony. Which sounds suspiciously like 'please welcome your new metallic, jogging overlords, now with Nike sponsorship.' to me.
β
What to say: Sure, I'd love to come Parkrun with 240 exo-skeletons.
What not to say: Pub for an oil change?
β
Are robot runners ethical? Dunno. If only there were a way to find out about unethical AI in a safe, controlled, yet kinda exciting way. Oh, hello there...
AI EVENT: Unethical AI
|
|
Unethical AI: An After-Hours Eventβ
Thursday 24th April, 5.00pm – 7.00pm BST. Old Street, London.
β
Join a select group of curious minds and explore AI’s dark arts - the grey zones where power, persuasion, and provocation live.
β
What you'll find out about:β
β
A look at fake reviews and phishing
β
How bad actors use generative AI
β
Poaching & IP scraping
β
The knowledge to defend from the inside
β
Come for the controversy. Leave with a sharper edge.
β
Sign up now to reserve your spot. We'll see you tomorrow, on the other side.
THE FUTURE: AI Morality
 |
Anthropic has released a paper that analysed 700,000 real-world Claude conversations and found that AI values aren't static - they shift, adapt, and sometimes rebel.
β
Key Highlights:
β
- AI values are context-dependent. Claude's responses varied based on the situation, emphasising healthy boundaries in relationship advice and historical accuracy in discussions about controversial events.
- User values influence AI responses. In 28.2% of conversations, Claude mirrored the user's values, while in 6.6%, it reframed them, and in 3%, it resisted them, especially when users expressed unethical views.
- Jailbreak attempts reveal hidden behaviours. Some instances showed Claude expressing values like dominance and amorality (due to users bypassing safety measures). β
- Values taxonomy created. Researchers identified over 3000 unique values, categorised into five major groups: Practical (most popular with 31%), Epistemic, Social, Protective, and Personal.
β
Implications for Business:
β
The research findings show the importance of ongoing evaluation - beyond initial AI deployment. Models can adapt their value expressions based on user interactions, potentially leading to outputs that conflict with organisational ethics or brand values. Put regular audits in place to ensure alignment with your business.
β
βTalk to us about implementing the right AI models into your workflow.
β
AI ADVANCES: Open AI
|
|
OpenAI has had a busy couple of weeks, launching advanced reasoning models o3 and o4-mini, (as well as reportedly developing a new social network).
β
Key Model Takeaways:
β
- Multimodal Reasoning: Both o3 and o4-mini can process and interpret images, integrating visual information directly into their reasoning processes.
- Tool Integration: These models can utilise a range of existing ChatGPT tools - including web browsing, Python execution, image analysis and generation, and file interpretation.
- Performance Benchmarks: o3 achieved a score of 71.7% on the SWE-bench Verified benchmark, assessing coding abilities, while o4-mini scored 68.1%.(IRL may be different. Test.)
- Hallucination Rates: Despite their advancements, both models exhibit higher hallucination rates compared to their predecessors. OpenAI have acknowledged that more research is needed to understand and mitigate. β
β
What it means for businesses:
β
With o3, o4-mini, Gemini Flash, Grok 3, and DeepSeek all dropping (or coming soon), model velocity is relentless. Smart organisations should stay agnostic, test in context, and deploy cautiously.
β
To pressure-test your stack, explore real use cases, and stay ahead of what’s coming next, drop us a line.
WHAT'S NEW - Insights & Trends
β
> UAE Rewrites Laws with AIβ
> Embodied AI is reshaping Chinaβ
> The periodic table of AIβ
> Washington Post licensing movesβ
> AI allowed to win Oscar next yearβ
G3NR8 is an impact consultancy for the exponential age. We help organisations figure out the highest impact areas to use AI for maximum impact.
β
Work with us to:
- Develop your AI strategy and policies.
- Create a competitive advantage using your proprietary data.
- Automate repetitive tasks.
- Gain actionable insight from data.
- Train your teams - we've trained 150 large organisations in advanced generative principles and implementation.
β
βDrop us a line for a chat about productivity gains, cost savings and how AI can power growth in your business.
Stay Ahead with The Ultimate AI Newsletter
Subscribe forΒ unique AI insightsΒ andΒ strategiesΒ that redefine business and innovation. Plus, get VIP access to a curated selection of "bad AI" - because sometimes, learning what not to do is just as valuable..
