Friday, August 22, 2025
Read time: 5-6 min
Read this article online
Hi, it’s Chad. Every Friday, I serve as your AI guide to help you navigate a rapidly evolving landscape, discern signals from noise and transform cutting-edge insights into practical leadership wisdom. Here’s what you need to know:
1. Sound Waves: Podcast Highlights
Next week, we’re celebrating our podcast’s one-year anniversary with a guest who perfectly captures why we started this show.
95% of middle-market companies are stuck doing AI busy work instead of seeing real ROI. That’s the uncomfortable truth Heather Murray dropped on our podcast this week. As founder of AI for Non-Techies, she’s generated over $75 million for clients including Toyota and Google while consulting for the UK government. Monday’s anniversary episode reveals her MICE framework for escaping “augmentation mode” and finding transformational returns. If you’re still measuring AI success by how much faster you complete tasks, this episode will restructure your Monday morning planning.
Subscribe for free today on your listening platform of choice because if you’re still in augmentation mode, this anniversary episode will change your Monday morning planning.
New episodes release every two weeks.
2. Algorithmic Musings. AI Plateau or AI Evolution? What Business Leaders Need to Know
The headline from August 14th read “AI Revolution Hits a Bump: GPT-5 Sparks Concerns of Plateauing Progress.” Uh oh. Perhaps I should sell my NVIDIA stock… But wait! One day later this statistical compilation boldly declared there to be no end in sight.
Here’s what’s actually happening.
If you’re a business leader trying to navigate the AI landscape right now, you’re probably feeling a bit like you’re watching a tennis match. Headlines volley back and forth between “AI is plateauing!” and “AI growth continues to explode!” Meanwhile, you’re sitting in the stands wondering: Which serve should I believe?
Both narratives contain elements of reality, but neither tells the complete story. What’s actually happening is far more nuanced and far more important for your organization’s future.
The Great AI Debate: Revolution or Evolution?
Picture AI development like your favorite 1990s rock band. (Stay with me here.) In the early days, every album was a massive leap forward. Nirvana went from Bleach to Nevermind. But eventually, even the most innovative bands settle into a pattern of steady evolution rather than revolution. That doesn’t mean they’ve stopped being creative or valuable. It just means the nature of their progress has changed.
That’s exactly what’s happening with AI right now.
What the “Plateau” People Are Seeing
The plateau narrative isn’t entirely wrong. Pure scaling is showing real limits.
Diminishing returns on pure scaling. Remember when everyone was obsessed with making models bigger and bigger? Well, it turns out you can’t just throw more compute at a problem forever and expect exponential improvements. GPT-4 used 70 times more computational power than GPT-3 but delivered only about 6 times the performance improvement. That’s still progress, but it’s not the hockey-stick growth we saw in earlier years.
Benchmark saturation. Many traditional AI benchmarks are getting maxed out. When models are scoring 90%+ on tests like MMLU, there’s simply less room for dramatic improvement. It’s like trying to shave seconds off a world record—the closer you get to perfection, the harder each incremental gain becomes.
Economic reality checks. Training frontier AI models now costs hundreds of millions of dollars. OpenAI reportedly lost $5 billion in 2024, largely due to training and inference costs. Even tech giants have to ask themselves: “Is the next 2% improvement worth another billion dollars?”
What the “No Plateau” People Are Seeing
While pure scaling might be hitting diminishing returns, innovation is accelerating in new directions:
Architectural breakthroughs are exploding. Instead of just making models bigger, researchers are making them smarter. Mixture-of-Experts architectures, retrieval-augmented generation, and small language models are achieving remarkable results with a fraction of the computational cost. DeepSeek-V3, for example, matches GPT-4’s performance at one-twentieth the training cost.
New benchmarks reveal massive headroom. While old tests are getting saturated, new challenges show AI still has enormous room to grow. On cutting-edge benchmarks like FrontierMath, even the best models solve only 2% of problems. On BigCodeBench, success rates hover around 35%. We’re nowhere near the ceiling.
Investment continues to pour in. If AI were truly plateauing, would venture capital firms have invested $122 billion in AI startups in the first half of 2025 alone? Would OpenAI have raised $40 billion? Smart money doesn’t typically chase plateauing technologies.
So What’s Really Happening?
AI isn’t plateauing. It’s maturing.
We’re transitioning from the “throw more compute at it” phase to the “work smarter, not just harder” phase. This isn’t a plateau; it’s an evolution from brute force to elegance.
Think about it this way: The automotive industry didn’t plateau when manufacturers stopped making cars dramatically larger and heavier. Instead, they focused on efficiency, safety, and user experience. The result? Today’s cars are faster, safer, and more efficient than ever without being gas-guzzling behemoths.
What This Means for Your Organization
If you’re making strategic decisions about AI adoption, this shift from revolution to evolution has three critical implications:
1. The Window for “Wait and See” Is Closing
During revolutionary phases, it sometimes makes sense to wait for the dust to settle. But during evolutionary phases? The technology becomes stable enough for real-world implementation while still improving steadily. Companies that move now get the advantage of the learning curve plus continuous improvements.
Are you still waiting for AI to “mature” before implementing it in your organization? That train has left the station.
2. Focus on Application, Not Just Capability
When AI was advancing by leaps and bounds every six months, it made sense to focus on raw capability. Now, the competitive advantage comes from how you apply these capabilities to solve real business problems.
The question isn’t “What’s the most powerful AI model?” anymore. It’s “What’s the most effective way to integrate AI into our specific workflows?”
3. Efficiency Matters More Than Ever
The shift toward smarter architectures means you don’t need the biggest, most expensive models to get substantial value. Small language models and specialized applications often deliver better ROI than frontier models for specific use cases.
This is fantastic news for medium-sized businesses. You don’t need OpenAI’s budget to benefit from AI’s capabilities.
Three Questions Every Leader Should Ask Right Now
As you navigate this evolving landscape, consider these critical questions:
1. Are we focusing on AI capabilities or AI applications? If you’re still primarily thinking about what AI can do rather than what AI should do for your specific business, you’re behind the curve.
2. What problems are we trying to solve? The most successful AI implementations start with clear business problems, not cool technology. What specific challenges in your organization could benefit from AI assistance?
3. How are we building AI literacy across our organization? The plateau narrative often comes from people who don’t understand the current state of AI development. Investing in AI education for your team isn’t just smart—it’s essential.
The Bottom Line
Don’t get distracted by the plateau versus progress debate. AI development is entering a phase of sustained, practical evolution that’s actually more valuable for most businesses than the previous era of dramatic but unstable breakthroughs.
The companies that thrive in this new phase won’t be the ones with the biggest AI budgets. They’ll be the ones that most thoughtfully integrate AI capabilities into their core business processes.
The revolution may be slowing down, but the evolution is just getting started. And evolution, my friends, is where the real value gets created.
Ready to figure out how your organization can ride this evolutionary wave instead of getting swept under by it? Let’s talk about building your AI strategy for the world we’re actually living in, not the headlines we’re reading about.
Want to explore what AI evolution means for your specific industry or organization? Drop me a line—I’d love to help you separate the signal from the noise.
3. Research Roundup: What the Data Tells Us
AI Hallucination Math: Why Chasing Perfect Accuracy is Burning Your Budget
I personally loathe the term “hallucination”. That said, we now have mathematical proof why we must stop chasing “hallucination-free” AI. New research from Samsung AI Center Warsaw demonstrates that perfect AI accuracy is literally impossible – not due to engineering limitations, but fundamental mathematical constraints.
The numbers that matter: Any AI system capable of meaningful work must violate at least one of four essential properties: truthful responses, complete knowledge use, full information revelation, or optimal output generation. The researchers proved this using three separate mathematical frameworks – this isn’t a maybe, it’s mathematical law.
The breakthrough insight: Hallucination and creativity are mathematically identical phenomena. The only difference is whether humans find the output useful. This means trying to eliminate hallucination entirely could kill the creative capabilities that make AI valuable in the first place.
What this means for your Monday morning: If you’re still demanding vendors deliver “zero hallucination” AI, you’re asking for something that violates the laws of mathematics. That budget you allocated to eliminate AI errors? Redirect it toward managing them strategically instead.
The catch: You still need accuracy for critical applications – just stop expecting perfection and using the imperfection as “evidence” that AI isn’t ready for prime time. Different use cases require different trade-offs between accuracy and capability.
Action item: This week, audit any AI vendor contracts or pilot requirements. If you see language demanding “zero hallucinations” or “100% accuracy,” you’re either paying too much or getting unusable systems. Replace these with specific error tolerance levels for each use case. Customer service chatbots? 95% accuracy might be fine. Financial calculations? You need 99.9%. Legal document review? Build in human verification workflows instead of demanding impossible perfection. Your next move: Email your procurement team this insight before your next AI vendor meeting.
Read our full analysis of this and all other analyzed research papers at AI for the C Suite.
4. Radar Hits: What’s Worth Your Attention
OpenAI’s progress page shows GPT-5 responses that sound remarkably human. This matters because we’re past the “AI sounds robotic” phase. If your team is still hesitating on AI pilots because the technology “isn’t ready yet,” you’re operating on outdated assumptions. The question isn’t whether AI can handle sophisticated tasks – it’s whether your competitors are already using it.
ChatGPT’s mobile app hit $2 billion in revenue since 2023. Translation for executives: consumers are paying $2.91 per download for AI tools, proving there’s real value being delivered. If you’re building customer-facing AI features, this validates that people will pay for quality AI experiences – but you’re competing against a very high bar.
Microsoft’s new COPILOT function brings AI directly into Excel formulas. This means your finance team can classify customer feedback, generate executive summaries, and automate trend analysis without leaving their spreadsheets. No more exporting to PowerBI or waiting for IT support. Action item: Ask your CFO when they last spent an hour manually categorizing data that Copilot could handle in minutes.
5. Elevate Your Leadership with AI for the C Suite
Ready to build an AI strategy for the world we’re actually living in, not the headlines we’re reading about? Let’s talk about how your organization rides this evolutionary wave instead of getting swept under by it.
As we navigate this unprecedented fusion of human and machine intelligence, remember: the best leaders aren’t just adapting to change – they’re actively shaping it. Until next week, keep pushing boundaries.
Chad