Bad AI: Trial By GPT

AI IS ACCELERATING - GET READY

Let's help you understand the WTF of AI and how it helps YOU and your organisation grow. Ready, set, agent.

This week:

😱 Bad AI: Trial By GPT

💭 AI Moves: ChatGPT Images

🔥 Elite Support: AI Training

✨ AI Scaling: Moore's Law Goes AI

💡 Trends & insights

BAD AI: Trial by GPT

​

This week a Norwegian chap found out ChatGPT had written him into a true crime novel.

​

According to our friendly neighbourhood neural net, he’d deleted his own children (like, permanently deleted them) and got 21 years in chokey for his troubles.

​

Plot twist: he didn’t. He’s alive, his kids are alive, everyone’s alive. Phew.

​

So what happened? Did the bot take a stand against the hirsute? Against fjords? IKEA? Who knows. What we do know is it pursued the claim with the confidence of a Daily Mail intern possessed by Skynet.

​

He’s now filed a GDPR complaint and unplugged Norway from the internet. (That last bit may need fact-checking, tbf.)

​

What not to say: C'mon it's only a hallucination. I've had worse microdosing.

What also not to say: Shall we give it a crack at running HR?

AI MOVES: ChatGPT Images

​

OpenAI has just pushed the button on a significant enhancement to ChatGPT: native image generation powered by their GPT-4o model. This allows users to create (much better) images directly within the chat interface - see the above image created from a simple 'make an image like this' prompt.

​

Key Highlights:

​

• Integrated Images: Users can now generate images within ChatGPT across all subscription tiers, including Free, Plus, Team, and Pro.
• Enhanced Accuracy: GPT-4o offers improved attribute binding, ensuring more precise and reliable image outputs. ​
• Advanced Text Rendering: The model is better at embedding coherent text within images, a notable advancement over previous versions using Dall-E. ​
• Autoregressive Approach: GPT-4o generates images sequentially, contributing to better text and character/image holding capabilities, albeit with slightly longer generation times.

​

Implications for Business:

​

For organisations leveraging AI for content creation - such as marketing materials, product designs, or educational content - the new model could streamline workflows by consolidating text and image generation into a single platform.

​

It's important to assess how this integration aligns with your specific operational needs and to be mindful of the model's speed, transparency - and cost.

​

We help companies in evaluating and implementing AI tools tailored to their requirements talk to us about what tools and models use and why.

AI SUPPORT: Train Your Team

​

AI is transforming how businesses operate - don't let your team fall behind. Invest in practical, high-impact AI training to ensure your organisation stays competitive in 2025 and beyond.

​

What you get:

​

✅ Up to 100 team members can attend.
✅ Pre-session survey and analysis to tailor training to your business.
✅ 90-minute interactive learning session.
✅ Follow-up AMA to address specific challenges and reinforce learning.

​

Investment: £3,500.

​

Companies that are strategically adopting AI are already gaining a competitive edge. Secure your session now and empower your team with the knowledge they need to succeed. Get in touch to book.

AI Scaling: Moores Law Goes AI

As AI gets more powerful, a key challenge remains: can it actually stay focused and finish real-world projects - not just answer short prompts? METR (Model Evaluation and Training Research) just dropped a fascinating studythat reframes how we evaluate AI: by the length of task it can complete autonomously with 50% reliability.

​

Key Takeaways:

​

• Time as the Benchmark: METR tracks how long a task takes a human expert and tests whether frontier models can complete it. The result? Success rates plummet as tasks stretch beyond a few minutes.
• Claude Sonnet 3.7 Hits 1 Hour: Today’s top models can now complete tasks that take humans around an hour - but they still fail on most tasks over 4 hours long.
• Exponential Progress: The length of tasks AI can handle has been doubling every 7 months for 6 years. If that holds, models could reliably take on week-long projects by 2027—and month-long ones by 2029.
• Real-World Stakes: This metric helps explain the disconnect: AI's seem superhuman on benchmarks, but struggle with the sustained execution needed for real work.

​

Why It Matters for Business:

​

For teams exploring AI for automation - think onboarding flows, research projects, or complex ops - this research is a useful. It’s not just about machine smarts; it’s about stamina. Before deploying agents into multi-step workflows, ask: how long can your AI actually stick with it?

​

We help companies build smarter, more durable AI workflows. Get in touch to assess which models match your task horizons.

WHATS NEW - Insights & Trends

​

> Google launches Gemini 2.5​

> Midjourney Creativity Paper​

> Bad candidates due to AI

> Perplexity bids for TikTok​

> Apple watch AI on the way​

G3NR8 is an impact consultancy for the exponential age. We help organisations figure out the highest impact areas to use AI for maximum impact.

​

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.

​

​Drop us a line for a chat about productivity gains, cost savings and how AI can power growth in your business.