Unless you’re Tom Cruise in a Mission Impossible movie, chances are, you’re not going to jump on a bullet train racing down the tracks at break-neck speed!

At the very least, perhaps let it slow down enough for you to safely jump on board? (Yes, I know it’s less stylish, but hey, at least you’re alive!)

No, this is not the premise for a new summer blockbuster hit.

Here’s the reality: AI isn’t racing ahead as many expected. Recent reports show that advancements in large language models (LLMs) are losing steam.

Let’s look at a few points.

OpenAI’s Project Orion

OpenAI’s much-anticipated Project Orion was expected to make waves, but it’s making ripples instead. Reports suggest modest improvements over GPT-4, with some surprising setbacks.

“Orion’s improvements over GPT-4 are less significant than the leap from GPT-3 to GPT-4. Notably, Orion struggles with certain tasks, such as coding, compared to its predecessor, although it excels in general language tasks like document summarization and email generation”[Source].

While Orion excels at some tasks, its struggles with coding highlight that progress isn’t always straight-forward… or easy.

Google’s Gemini Challenges

Google’s Gemini model, another industry heavyweight, isn’t immune to the slowdown.

“Google’s AI development is experiencing lagging improvement rates compared to the previous model versions. Past models exhibited faster rates with more training data and computing power. However, the current AI model is improving at a slower pace, even when using more processing data and specialized AI chips”[Source].

Even the tech titans aren’t immune to hitting plateaus, proving that scaling AI isn’t just about throwing more data and compute power at the problem.

Industry-Wide Trend

The slowdown isn’t isolated to a few companies. Bloomberg puts it plainly:

“Three of the leading AI companies are now seeing diminishing returns from their costly efforts to build newer models”[Source].

This trend is reshaping how businesses think about AI and its next steps.

Expert Opinions

What do the experts think? Turns out, they’ve been predicting this for a while:

• Gary Marcus from New York University: “The outrageously high valuations of companies like OpenAI and Microsoft are largely based on the idea that LLMs, as they continue to scale, will become general artificial intelligence… this is just a fantasy”.
• Sasha Luccioni, a leading researcher in ethical artificial intelligence: “The pursuit of AGI has always been unrealistic, and the ‘bigger is better’ approach to AI was bound to hit a limit eventually—and I think that’s what we’re seeing now”.
• Scott Stevenson, CEO of Spellbook: “Some labs were very focused on just supplying more languages, thinking that AI would just get smarter”.

Perhaps before we decide if this is the train we want to hop on, let’s look at the causes of the slowdown.

What’s causing the slowdown?

Why is AI Losing Steam?

Even the most powerful warriors from history, fantasy or mythology, will run up against enemies that test their strengths.

For AI, these are the Top 3 Big Bads.

Hurdle #1 – Computational Limits

Scaling AI models is hitting a wall, and the costs are astronomical:

• Exponential Demands: Creating a model 100x larger than GPT-4 would require a million GPUs—beyond the current infrastructure.

• Shrinking Gains: Scaling laws suggest we’re nearing the limit of what brute-force computing can achieve.

Hurdle #2 – Data Saturation

AI models thrive on data, but the well is running dry:

• Limited Domains: High-quality datasets for complex fields like reasoning are scarce.

• Data Restrictions: More websites and platforms are restricting AI’s access to their content, shrinking the pool of fresh training data.

• Plateau Effect: Ilya Sutskever, who recently left OpenAI to start SSI, calls this the “AI data wall,” where the lack of new data halts significant improvements.

Hurdle #3 – Hardware Constraints

Building and training larger models is constrained by hardware:

• GPU Shortages: Demand outpaces supply, stalling projects globally.

• Energy Consumption: AI training sucks up massive amounts of power, adding environmental and operational costs.

• Infrastructure Gaps: Companies must now focus on maximizing existing resources rather than just scaling up.

These three constraints acting together are stacking against the growth of AI.

So is this the end of the hype? Are we back to the pre-AI days?

My Master, in his infinite wisdom, had always told me that to plan for the road ahead, we need to look at the trail behind.

And believe it or not, this is not the first slow down we see.

This is not the first time tech development slowdown in history.

Even the Fastest Trains will Slow Down near “S” Curves

If you look back at the last hundred years (don’t worry, I already did), we would see a trend. 

You see, technology often evolves in S-curves—periods of rapid innovation, followed by plateaus, and then breakthrough shifts.

Here are a couple of examples you’d be familiar with:

1. Smartphones

• Rapid Innovation: From IBM’s Simon in 1994 to Apple’s iPhone in 2007, smartphones revolutionized communication and content consumption, creating a new global standard.

• Plateaus: Over the past decade, innovation has become incremental. Features like better cameras and marginally faster processors dominate, but revolutionary advancements are rare.

• Breakthrough: Emerging technologies like foldable screens, AR/VR integration, and AI-powered applications hint at the next phase of smartphone evolution.

2. Steam Engines

• Rapid Innovation: During the Industrial Revolution, steam engines transformed transportation and industry, driving unprecedented growth and efficiency.

• Plateaus: By the 1920s, steam engines reached their peak efficiency, unable to push past their physical limitations.

• Breakthrough: The rise of electric motors and internal combustion engines disrupted the steam engine’s dominance, leading to a new era of energy and mobility.

3. Internet and Web Technologies

• Rapid Innovation: The late 1990s saw an explosion of innovation during the “Browser Wars,” where rapid advancements turned browsers into interactive, programmable platforms.

• Plateaus: Modern browsers now focus on refinement—security enhancements, performance boosts, and user experience improvements—without transformative changes.

• Breakthrough: As the Internet of Things (IoT), Web3, and immersive web technologies emerge, we’re on the brink of another internet revolution.

So as you can see, AI is not dead. As a matter of fact, it has just gone through its fair share of Rapid Innovation and is now in the Plateaus stage.

You can say it’s the calm before the storm, the pause before the sprint, the receding waves before a tsunami.

But just as my Master always reminded me, sometimes you need to slow down to speed up, and when you pause more, you see more.

And being an AI opportunist, I actually see this as an opportunity!

There’s always an opportunity in every situation.

Why This Slowdown is Actually a Good Thing for Businesses and Organizations

If you’ve been on a road trip, sometimes it pays to stop by the road and check the map, the vehicle, or even just on the passengers.

And this AI slowdown is your opportunity to do that. And for some businesses, this could even be the best thing to happen!

1. A Chance to Hop Onboard

Imagine you’re late to the AI party.

All the big players seem lightyears ahead, and you’re stuck figuring out where to start.

The slowdown? It’s like the train pausing at the station, giving you just enough time to climb aboard. (So you don’t have to do that Tom Cruise stunt 🤣)

Opportunity for Late Adopters: I see this as a chance for businesses that missed the first wave of AI adoption now have the breathing room to catch up without falling further behind.

Plus, even if you’re a small or mid-size company, you can now explore tailored AI tools without competing against cutting-edge, fast-moving giants.

2. Time to Breathe and Rethink

When the pace slows, there’s a rare chance to step back and ask:

“Are we doing this right?”

Strategic Realignment: Use this moment to revisit your AI goals. Are you just chasing shiny tech, or are you solving real problems?

Experimentation Space: Try lightweight AI models, pilot domain-specific applications, or test ideas before committing to big budgets.

Case Study: Nike’s e-commerce pivot during COVID-19 is a perfect example. Instead of panicking, they leaned into building an online retail channel, which became a significant revenue driver. The rest, as they say, is history (although if you really think about it, it was barely 3 years ago…#timeflies)

3. A Moment to Course-Correct

Let’s be honest: moving fast sometimes means moving in the wrong direction, often without knowing it until it’s too late. The slowdown lets you pause, check your map, and adjust course.

Course Correction: Evaluate whether your AI initiatives are genuinely aligned with your broader business strategy.

Risk Reduction: Spot potential pitfalls like biased AI models or inefficient systems before they spiral out of control.

Expert Insight: Erik Brynjolfsson’s “productivity paradox” shows us that tech slowdowns often precede massive breakthroughs. Use this lull to prepare!

4. Build a Stronger Foundation

Think of this slowdown as a construction phase.

It’s time to reinforce your base before the next big wave hits.

Upskilling Teams: Train your employees to work alongside AI tools effectively. Build the literacy now, so you’re not playing catch-up later.

Infrastructure Development: Invest in systems like data lakes or integration frameworks to ensure you’re ready for future advancements.

Case Study: Walmart’s Data Café initiative is a shining example. By processing data from over 20,000 stores, they enabled faster decisions and better ROI.

Capacity Building. Systems Building. Team Building.

What Businesses and Organizations Can Do During This Slowdown 

If you’ve followed any of our #bebrilliantwithAI articles before, you’d know no article is complete with Actionable Insights.

The AI race might have slowed, but that doesn’t mean we at Radiant Institute are.

As a matter of fact, we want you to run faster, which is why we’ve lined up a few recommendations for you.

Recommendation #1 – Build Internal Muscle

Capabilities-Building: Teach your teams how to use AI tools effectively. AI literacy is no longer optional—it’s essential.

Example: Follow Walmart’s lead. They invested heavily in internal data management, and it paid off with faster decision-making and improved efficiency.

Recommendation #2 – Test Small, Win Big

Proof of Concept (POC): Start small. Test AI capabilities in bite-sized projects before scaling up.

Example: Cisco’s approach in India highlights the value of POC projects to validate ROI without overcommitting.

Recommendation #3 – Stay Laser-Focused on Goals

Strategic Alignment: AI isn’t a magic wand. Tie your AI initiatives directly to business outcomes—whether it’s cutting costs, improving efficiency, or delighting customers.

Example: Procter & Gamble’s adoption of AI in product development and supply chain optimization significantly improved efficiency and reduced costs. That’s ROI done right.

Recommendation #4 – Future-Proof Your Systems

Invest in Data Infrastructure: Build systems that can handle advanced AI tools down the line. Think of it as setting the stage for Act II.

Example: Nestlé’s use of AI for predictive analytics has transformed their supply chain, allowing them to process vast amounts of data to optimize operations and improve decision-making. If they can do it, so can you.

Recommendation #5 – Stay Nimble

Adapt and Pivot: The AI landscape is evolving fast. Avoid locking yourself into rigid, long-term tech commitments.

Expert Advice: Pareekh Jain, a leading analyst and founder of Pareekh Consulting, reminds us that flexibility is the secret to thriving in uncertain times.

Thinking out-of-the-box.

Innovation Comes During Slowdowns

The AI slowdown isn’t a roadblock—it’s an open door.

It’s a call to rethink how we approach challenges and to optimize what we already have.

And I thought I’d end today’s article with what might look like a breakthrough to this AI slowdown problem.

Let me close with the story of Test Time Compute (TTC).

For the uninitiated, TTC is a method that emphasizes smarter use of resources during the deployment phase of AI models rather than relying solely on massive pre-training efforts.

It shifts the focus from brute force computation to strategic optimization, enhancing performance by allowing models to prioritize reasoning and efficiency in real-time.

In other words, TTC encourages AI to work smarter, not harder.

TTC is considered an innovation because it’s kinda coloring outside the lines. By rethinking traditional approaches, TTC offers a breakthrough in addressing the challenges of computational limits. It demonstrates that slowing down, re-evaluating processes, and focusing on deliberate adjustments can lead to meaningful advancements.

This innovation is a lesson in adaptability, showing that less can indeed be more when resources are used wisely.

While still in it’s early stage, TTC teaches us a powerful lesson: you don’t need more; you need better. Instead of piling on more data or horsepower, TTC focuses on smarter, optimized use of resources.

Businesses can learn from TTC by optimizing existing processes and aligning resources more strategically.

In other words…Strength in Deliberation. Tough times don’t last, but tough people do.

Since we started this article talking about trains, let’s end it with a boat.

Particularly, Terry Barkman’s Sailboat Principle, which perfectly sums up our article – slow, deliberate coordination often leads to faster, more effective outcomes.

AI adoption is no different. This slowdown is your moment. Use it to recalibrate, build, and prepare for the next wave of innovation.

💡 Remember: it’s not about how fast you go; it’s about going in the right direction.