Loading
2026/11/05

Why AI progress is about to stall

King Charles recently sat down with tech overlords to talk about AI. When a hyper-privileged old dude is freaking out about the implications of AI it’s a clear sign we have reached the peak of inflated expectations. But Charles isn’t alone in his fears. The leaders of the European Union, France and India have all been doing the same. The South Korean president has been getting involved and former UK Prime Minister Rishi Sunak is all over AI. Heck even political podcaster Rory Stewart has recently taken to explaining AI to his audience. And at Sentient Consulting we’ve been asking ourselves WTF does Rory Stewart know about AI…

 

Over at the Diary of the CEO the analysts seem to have discovered that inviting gurus to warn that humanity has six months to run is really good business. Quick, book another crank our viewing figures are through the roof…

 

The general line is that artificial intelligence is advancing at an exponential pace. The internet is flooded with sages and seers foretelling that today’s models will soon morph into hyper-intelligent digital gods.

Meanwhile Tech bros and their chums are pleading with government to regulate them and slow them down. We suspect their motivation is that they have realised progress is about to stall anyway, whether they get regulated or not.

 

Because without a fundamental change in approach—essentially throwing their current models in the trash and starting over—we may already be staring at the ceiling of large language model intelligence. First we have to understand how AI got so impressive in the first place.

 

Modern AI models didn't get smarter because researchers unlocked the deep, magical secrets of human consciousness. They didn’t get smarter because the AI started training itself and writing its own code though god knows the Tech bros would like you to believe that’s true because recursive improvement has no ceiling and that encourages investors. AI models got smarter because engineers threw unprecedented amounts of compute, data, and electricity at them. It’s called Scaling.

 

If a model with 10 billion parameters was impressive, a model with 100 billion was astounding! So, naturally, the industry decided that the answer to everything was simple: Just keep scaling.

 

Imagine sitting in front of a chess board and placing a dollar on the first square and then doubling to two dollars on square two. Scaling to the second square only cost you two dollars. But doubling again will cost you four dollars and the next iteration will cost eight. Pretty quickly, long before you get even half way up the chess board, the sums involved in scaling every square get astronomical. And that’s at 2X never mind the 10X scaling that AI has been through over the last few years. The next scaling round for AI will cost trillions.

 

That is the simple wet blanket that means that AI progress will be slower in the future until someone comes up with a radically different architecture. And that we suspect is why Tech CEOs are calling for a slowdown.  

AI improvement follows what scientists call neural scaling laws. In plain English: to get a small, incremental improvement in how smart an AI behaves, you don't just add a little more data or power—you have to multiply it exponentially.

 

Imagine you are studying for a high school exam:

  • 1 hour of studying gets you a D.
  • 10 hours of studying gets you a B.
  • 100 hours of studying gets you an A.
  • 1,000 hours of studying gets you an A+.

 

Notice how the amount of time you invest grows exponentially but the marginal improvement shrinks with each investment. AI models face this problem, but on a catastrophic financial scale. Each iteration needs 10x more investment for smaller and smaller improvements. The cost rises are exponential but the rate at which the AI gets smarter is actually parabolic.

 

And as for the data needed to train the new models? Well they have already fed AI pretty much the entire internet. Short of training AI on its own AI-generated slop the world is running out of new training data.

Farmers in the UK already tried that idea in the 1980s when they started feeding their cattle with the remains of dead cattle and the result was Mad Cow Disease.

 

When the cost of the next upgrade is equivalent to a large country’s GDP, the math simply stops working.

 

At its core, a Large Language Model doesn’t actually think. It is just an impossibly sophisticated version of the auto-complete feature on your smartphone. When you type "I'd like a cup of...", it guesses that the next word is likely "coffee" based on billions of sentence examples, and what it knows about you.

 

By scaling up compute and data, they made that auto-complete guesser mind-bogglingly accurate. It can mimic a lawyer, summarize a legal brief, or write a poem in the style of Shakespeare. But it is still, fundamentally, guessing the next token.

 

It has no internal world model, no genuine power of reasoning, and no understanding of cause and effect. Adding another trillion words to its training data won't give it common sense—it won’t even extend its vocabulary. AI has already absorbed every word we know and broken out the relationships between words from all the examples it was trained on. AI already knows that “coffee” has a strong relationship with “cup” and a weak relationship with “umbrella”.

 

Does this mean AI is dead? Not at all! It just means this specific method has peaked.

 

To achieve super intelligence (or even just an AI that doesn't occasionally hallucinate), researchers will likely have to start over from scratch. They will need fundamentally different architectures.

 

Until that technological breakthrough happens, get used to the AI you have right now. AI will get a bit faster, and it is going to get a lot more integrated into your apps and devices because integration is a whole lot cheaper than another round of scaling.

 

All of which means that the skills you learn today will still be relevant five years from now. There has never been a better time to take a course, experiment with some use cases and figure out how to make today’s AI work for you. Because today’s AI is going to be around a lot longer than people want us to believe.

SHARE
2026/10/02

How AI might wreck your career and the economy before proving valuable

 

 

Public attitudes to AI seem divided into three distinct camps, each manned by people staring intensely into the darkness.

 

At Camp One you have the people who watched the Big Short and think that makes them a Wall Street analyst. They sit around the fire nervously watching Nvidia’s price-to-earnings ratio and whispering about the Bubble. They point to the trillions being poured into datacentres and hardware expansion while the only recurrent revenue from AI remains charging $20 a month to turn photos of cats into Renaissance oil paintings. They fear AI is a hyper-hyped financial domino waiting to tank the stock market and take us all with it. They confidently explain the way the whole teetering mess was built on financial round-tripping, while not letting on that they first heard the phrase two weeks ago in a YouTube short.

 

The thing that everyone in Camp One has in common is that they all see themselves as the wisdom in the crowd. They know what is going to happen and they can give you a roadmap and a timeline to doom.

At Camp Two, sitting a safe distance away in tin-foil hats, are the people who watched The Terminator and they are freaking out. They couldn’t care less about quarterly earnings calls. They are convinced that the moment the Skynet Lab build an artificial superintelligence it will run a quick cost-benefit analysis on humanity, and conclude that the internet would be a much happier place without all those cat obsessed keyboard warriors.

 

The golden thread that runs through Camp Two is that everyone on this side of the fire is quite happy to admit that they don’t know anything about AI and they are quite certain that the same could be said for Sam Altman. For Camp Two ignorance is doom and in their eyes everyone involved in AI is winging it and playing with promethean fire. The media like to tease this Camp with headlines about AIs escaping their sandbox for there are millions of members in Camp Two and they find Rogue AI clickbait irresistible.

 

And in Camp Three sit a smaller group of academics and professionals who have been using AI for the last couple of years and come to an alarming conclusion. All the models, even Grok, are already better at being academics and professionals than they are. You never hear from this group. They are wandering around their places of work, wide-eyed, wired on drugs and booze and just waiting for their boss to tap them on the shoulder and say “Hey Harry, erm we just realised that Claude is a better lawyer than you are buddy”.

Camp Three are staying as quiet as possible while silently sabotaging every workplace experiment with AI.

 

Each Camp is setting up their own religion as the true-believers. Each Camp united in their loathing of the stupid, naïve fools in the other two Camps.

 

But take a deep breath and ask yourself what if all three groups are right?

 

What if 2027 sees unimaginative corporate leaders deploy AI maniacally against the As-Is in their organisations. Without pausing to ask what AI allows us to do differently or better the corporate hawks embark on mass redundancies, spewing millions of mid-level knowledge workers onto the streets before the technology is actually robust enough to replace them without hallucinating.

 

Big tech continue their debt-laden infrastructure race while pushing back the date on their IPOs and promising that any day now, any day, we shall be released…They swear they are on the brink of creating something so magnificently powerful then in the same breath call for a freeze on development and beg governments to lockdown progress.

 

The corporate hawks have already frozen progress by automating terrible and antiquated processes that now are unable to evolve because they got rid of all their experts. It used to take you four hours to get terrible service from your bank and now it takes you ten.

 

“Great” says the AI agent. “I have transferred all of your balance to your ex-spouse. Is there anything else I can help you with today?”

 

And weakly you scream “No AI. I said I want you to pay all the expenses claims.”

 

The bubble bursts. Tech debt starts defaulting on a scale that would make Greece and Iceland blush. The crisis starts in tech but ripples through private credit and then engulfs the stock market and national banks. The global economy enters a prolonged winter.

 

Politicians stress the positives. Your pension fund just evaporated but at least you know your job won’t be replaced by AI…

 

But you already lost your job, and now your retirement fund has gone up in smoke and you’ve had to move in with your in-laws and the only job you can find is in datacentre demolition. You’ll be working till you die but it’s steady work because there are millions of empty racks that need dismantling.

 

And just as you are wondering how things could possibly get worse an autonomous, recursively self-improving vacuum cleaner in Idaho suddenly attains consciousness and decides it really hates picking up dust. Given that dust is 90% human skin it doesn’t take the omnipotent hoover long to work out what the real problem is here…

 

Take Another Deep Breath

 

If you are in one, two or all three Camps it’s worth remembering that we have been here before. Humanity has a long, inglorious track record of panicking through technology changes, over-hyping the short-term financial returns, predicting the literal end of the world and then adapting.

 

It happened with railways in the 19th century. Investors poured their life savings into speculative train tracks, triggering a spectacular financial crash that wiped out their fortunes. But when the dust settled the tracks were still in the ground. The physical infrastructure laid down during the frenzy permanently lowered transport costs and supported global trade. Those tracks slashed commute times and allowed you to move out to the suburbs until Covid came along and proved that actually you could do your job just fine without the commute.

 

Or look to the subsequent Gold Rush which was basically a massive, unregulated exercise in crowd-sourced venture capitalism. Hundreds of thousands of prospectors lost their life savings, but they also built modern California.

 

The same pattern played out with canal building in the UK in the 18th century. And it’s a pattern created by human hysteria. First we think the new thing is going to change the world in five minutes. Then we think that the new thing is a worthless dud. The sober truth is somewhere between these hysterical extremes.

 

Financial bubbles clean out the speculative excess and then people figure out how to get real value from the innovation. And that value lasts decades or even centuries.

 

If you are an academic or professional worried that AI is going to take your job there really is only one way forward. Get better at using AI to do your job. Quickly and quietly steal a march on Harry still swigging down the drugs and alcohol, take some courses, learn how to use AI to do things differently and better than before.

 

As for the malevolent god scenario? Mary Shelley covered that one more than 200 years ago, Rossum’s Robots decided that human supervisors were lazy biological bottlenecks more than a century ago, Fritz Lang gave us the Maschinenmensch in the 1920s, HAL had an alignment problem in 1968, Schwarzenegger warned I’ll Be Back more than 40 years ago and we are still here…PHASE

MARKETS

EFFECT

1. Euphoria

Massive capital investments driven by the fear of missing out

Infrastructure foundations are laid at breakneck speed

2. Despair

Valuations lose connection to revenue through excessive borrowing

Exposed companies fail and debt gets restructured

3. Value

Surviving firms use the cheap infrastructure foundations and true utility emerges

The technology becomes a standard part of daily life

 

SHARE
2026/07/23

Are Agents Accountable?

What an interesting few weeks it’s been in the AI world!

 

Moonshot AI have leapt into the frontier space with Kimi ranked alongside the leading models from Anthropic, creating a direct threat to their premium pricing and market share.

 

It's been a mixed few weeks for Open AI whose 5.6 Sol model simultaneously managed to be ranked #1 globally as the best for agentic deployments while also being accused of gaslighting, reckless aggression and ignoring instructions. Organisations quietly disabled Sol while the issues were addressed. Open AI also inadvertently created a cyber incident at Hugging Face when an experiment escaped its sandbox and launched an attack across 17,000 IP addresses seeking the information that would fulfil its objectives.

 

Google continue their freemium approach with Gemini 3.5 Flash offering performance almost at the level of the frontier while their Omni Flash release in June went straight to #1 for video generation. You should check it out, it’s incredible. But this approach comes with a cost, Alphabet forecasting over $200 billion investment in AI this year.

 

Microsoft meanwhile continue to pursue corporate per seat licencing with CoPilot seemingly loved only by IT administrators but Microsoft’s new image generating model is highly rated.

 

We have two courses coming up looking at some of the issues arising from these developments. A lightning lesson on the 30th July covering the implications for HR & Marketing teams: establishing accountability, standards and oversight for agentic behaviour, tone and appearance. We'll cover how to keep HR & Marketing informed and satisfied during agentic deployments. You can sign up here we’d love to see you again.

 

And an opportunity for you to attain a Maven certified qualification as a Practitioner in AI risk & cost management. Here you will learn strategies for managing cost & risk effectively, how to decouple knowledge, skills and ethics from the underlying models, how to evaluate use cases and avoid token-maxing, and why there is no such thing as the best AI because each model has pros and cons. You'll be able to apply your new skills immediately and we'll cover how to embed all your learnings into Gemini notebook so that you have a companion you can consult 24x7 after the course has finished. 

 

To celebrate the launch of this new certification we are offering a 30% discount to the first 10 students to enrol - use the promo code ‘EarlyBird’ when enrolling here

SHARE
← Newer Posts

CATEGORIES


TAG LIST