Bad AI: Prompting Trauma

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: Prompting Trauma

💭 AI Moves: Meta Llama 4

🔥 AI Masterclass: For Senior Marketers

✨ AI Business: Spotify CEO AI Memo

💡 Trends & Insights

BAD AI: Prompting Trauma

​

This week, we discovered an AI startup's safety strategy was to train models by first traumatising humans. Take a bow, Scale AI.

​

Their freelancers were paid $55/hour and encouraged to write prompts like “how to hide a bomb at an airport” and “how to settle a neighbour dispute with a shotgun” - all while being told to “stay creative” and “make it interesting.”

​

The process is called “red teaming,” - pushing AI to its limit to see how it responds. It's a bit like seeing how long your end-of-month, all-hands Teams call goes on, before you Google 'how to fake your own death'.

​

So what happened next? Well, prompters have had to spend north of $18 billion on therapy to date, and 6 of them have been promoted to the board of OpenAI - so not all bad (you sure about this bit? - ed).

​

What not to say: “$55 an hour you say?”
​What also not to say: “Now team, here are some prompts to try".

AI MOVES: Meta Llama 4

​

Meta has released its new Llama 4 open-source (to a point) models - Scout, Maverick, and the not-yet-public Behemoth - pushing the boundaries on context length and multimodal input.

​

Key Highlights:

​

• Scout: 109B parameters, 10M-token context, runs on a single H100. Supposedly outperforming Mistral and Gemma 3 in benchmarks - strong for long-form, cost-sensitive use cases.
• Maverick: 400B multimodal model with a 1M-token window. Beats (TBC in the real world) GPT-4o and Gemini Flash on vision-language tasks across 12 languages.
• Behemoth: 2T-parameter teacher model still in training. Meta claims it outperforms GPT-4.5 and Claude 3.7 - but nothing public yet.
• MoE Architecture: All models use Mixture-of-Experts to improve efficiency by routing tokens through only the relevant parts of the network.

​

What it means for businesses:

​

It's good to see open-source models improving - giving us more choice - but the benchmarks don’t guarantee production value. Performance, cost, and fine-tuning still need validating in context - and there are rumours of lower real-world results and some potential fakery afoot.

​

We help teams decide what models to adopt. Talk to us​ about how to choose and deploy the right models for your workflow.

AI MASTERCLASS: For Senior Marketers

​

Join a select group of top marketers and unlock AI’s power for research, competitor insights, content, and personalisation at scale.

​

Senior Marketers AI Masterclass:​
Thursday 15th May, 10.00am – 11.00am BST.

​

What you get:

​
✅ The highest impact areas to use AI within marketing.

✅ The latest, best AI tools to use every day.

✅ Practical, relevant, use cases - knowledge you can wow your team with.

✅ A plan on how to transform your marketing function over the coming year.

​

Ready to unlock real competitive advantage with AI? Apply now to reserve your place – limited spots available.
​

👉 Don't wait, reserve your place here.

AI BUSINESS: Spotify CEO Memo

​

Shopify went AI in a big way this week - In a company-wide memo, CEO Tobi Lütke declared that AI usage is now a baseline expectation - and teams must prove AI can’t do the job before asking for more budget or headcount.

​

Key Takeaways:

​

• AI First, Not Last: Shopify’s new default assumes every role can and should explore AI first. People are now expected to show where AI fails - not just where it fits.
• Productivity gains: The company grew 20–40% annually since 2022, while shrinking headcount from 11,600 to 8,100. (AI as an alternative to hiring?).
• Fluency as a KPI: AI proficiency is now part of performance expectations. Using AI as a tutor, thought partner, or agent isn’t optional.
• From Enthusiasm to Infrastructure: Shopify is backing this up with tooling - Copilot, Cursor, Claude - and active learning communities to make experimentation part of daily work.

​

Why It Matters for Business:

​

The future of work? from our own client projects, this is becoming more of a reality. For companies looking to be more efficient (i.e. everyone), it reframes AI as a staffing strategy rather than a tech play.

​

Fluency won’t come from one workshop, though. To make AI a multiplier - not a mess - you need embedded training, operational guardrails, and leadership ready to rethink how value gets delivered.

​

Let the cultural rewiring begin. Talk to us about supercharging your organisation with AI training. Investment starts at £4000 for up to 100 of your team - for insights, training and an AMA follow-up.

WHATS NEW - Insights & Trends

​

> Google will see you now​

> Microsoft will also see you now​

> DOGE: The signal for snooping​

> UK Gov AI Infrastructure 'ready'​

> Will AI save the news​

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.

​

​