AI & NLP Enthusiast | Helps Early Founders Define Ideal Customers & Launch Successful GTM initiatives | Advisor | Growth Hacker | Fractional CRO

I build and scale high-performance GTM and enablement functions for technology companies.

As a hands-on leader, my mission is to drive predictable, scalable growth by applying a strategic, operator-first mindset.

Honcho Pickleball: Scaled GTM from 1 → 32 cities, driving 200% revenue growth in 18 months.

Metadata.io: As the first sales hire, created the company's GTM playbook and secured initial enterprise clients.

LivingSocial: Launched Amazon-backed GTM strategies in 10+ cities, training and scaling new sales teams.

I leverage AI automation and RevOps frameworks to build scalable systems that consistently deliver 3x ROI and boost sales efficiency.

If you are looking for a leader who can transform your commercial channel with a proven, results-driven approach, let's talk.

Chris M.

“the only metrics that will truly matter to my life are the individuals whom I have been able to help, one by one, to become better people.” ― Clayton Christensen

I build tools that help people think, not tools that think for people.

Founder & CEO of Curiouser.AI, a values-based AI company committed to human intelligence, imagination, and the rebirth of trust.

Instructor at UC Berkeley.

3x founder.

Harvard MBA.

My work sits at the intersection of AI, ethics, philosophy, and strategic imagination.

I write daily about what technology is doing to us, what it could do for us, and how leaders can navigate the age of acceleration without losing their humanity.

I believe the future belongs to those who slow down, think deeply, and design technology that elevates people instead of replacing them.

If you’re here for honest analysis, clarity, and a saner approach to AI, welcome. I’m glad you’re here.

(LinkedIn caps connections at 30,000, but please feel free to follow me. I’m grateful to stay in conversation through my writing.)

Stephen Klein

The Age of Pretend
The AI industry just spent $200 billion on the wrong problem.

Everyone assumes AI’s biggest bottleneck is compute.
More GPUs. Bigger clusters. The GPU arms race.

That assumption is wrong.

The real bottleneck isn’t compute.
It’s architecture, specifically, a design decision made in 1945.

Let me explain.

In a recent interview, Ilya Sutskever, co-founder of OpenAI and arguably the most important technical mind in modern AI, said this:

“I don’t think current hardware is a limitation. It’s just not the case.”

Read that again.

The person who helped build GPT is not worried about compute.
He’s worried about ideas.

The real constraint: the von Neumann bottleneck

Modern computers separate memory and processing.
Data has to move back and forth between them.

For most software, that’s fine.
For AI, it’s catastrophic.

Some numbers the industry rarely highlights:

• Accessing off-chip memory consumes ~200× more energy than the computation itself
• Roughly 80% of Google TPU energy goes to electrical connections, not math
• A 70-billion-parameter model moves ~140 GB of data just to generate one token

The computation itself?
Simple matrix multiplication. Modern chips are excellent at it.

The problem is data movement.

The scaling mismatch

Here’s the part that should stop everyone cold:

• Training costs for frontier models have scaled ~750× every two years
• Memory bandwidth has grown ~1.6×
• Peak hardware FLOPS increased ~60,000×

We’re building faster engines, and forcing them to drink through a straw.

This is why Sutskever declared the Age of Scaling effectively over.
From roughly 2020–2025, brute-force compute worked.

That era is ending.

“The next major improvements will likely come from smarter algorithms, not more expensive hardware.”

Why this matters

• Companies pouring billions into GPU clusters may be solving yesterday’s problem
• The next real breakthrough is likely architectural, not incremental
• Smaller teams with novel approaches may leapfrog massively funded incumbents

IBM Research is already exploring in-memory computing—processing data where it lives. Early results suggest 100–1,000× energy efficiency gains.

That’s not optimization.
That’s a paradigm shift.

Yet most AI discourse remains fixated on NVIDIA earnings calls, GPU supply, and capex arms races.

The industry made a classic mistake: confusing correlation with causation.

GPT-3 was big and good → therefore bigger = better.
Scaling worked → therefore scaling is the answer.

Until it isn’t.

We’re watching hundreds of billions flow into hardware while the true constraint is architectural thinking.
-
"THE TRICK WITH TECHNOLOGY IS TO AVOID SPREADING DARKNESS AT THE SPEED OF LIGHT"

Invest in AI 2.0 and own a piece of what comes next. We are raising on WeFunder (Link in comments)

Founder & CEO, Curiouser.AI | Berkeley Instructor | Building Values-Based, Human-Centered AI | LinkedIn Top Voice in AI

MIT AI courses are FREE right now (No coding required). 🤯
(You may want to consider sharing because LinkedIn is likely to step on this because they don't links that take someone off the platform and this a real PSA)

Many people are paying for “AI certifications”…

While MIT is giving real AI learning for free.

If you want to understand AI from basics to deep learning + GenAI, this list is gold.

Here are 9 FREE MIT AI courses you should take in 2026:

1) Introduction to Machine Learning
Learn how ML problems are framed, and how models learn through supervised + reinforcement learning (images + sequences).
🔗 https://lnkd.in/d6WyUQsx


2) Artificial Intelligence
Learn the core building blocks of AI: knowledge representation, problem-solving methods, vision, and language.
🔗 https://lnkd.in/dMxm7qW2


3) Foundation Models and Generative AI
Learn how GenAI really works, why foundation models changed AI, and how self-supervised learning made this possible (non-technical).
🔗 https://lnkd.in/d5-Dqfp4


4) AI 101
A beginner-friendly MIT workshop that explains AI concepts in simple terms, plus a hands-on exercise to train your own algorithm.
🔗 https://lnkd.in/dZ8VRsB6


5) Introduction to Deep Learning
Learn neural networks fast through a focused MIT bootcamp with applications in NLP, vision, biology, LLMs, and GenAI.
🔗 https://lnkd.in/dX7ESNqx


6) ML with Python (MITx)
Learn machine learning from linear models to deep learning + reinforcement learning using hands-on Python projects.
🔗 https://lnkd.in/dj228C79 


7) How to AI (Almost) Anything
Learn how AI works beyond text, across images, audio, sensors, music, art, and real-world multimodal systems.
🔗 https://lnkd.in/dhC4j6WU


8) Artificial Intelligence in K–12 Education
Learn GenAI fundamentals and how it impacts education, along with project-based learning focused on real classroom use cases.
🔗 https://lnkd.in/dRZcXuzk


9) Introduction to Algorithms
Learn how algorithms, data structures, and performance analysis form the base of fast and reliable AI systems.
🔗 https://lnkd.in/ddQqcDTh
-
Curiouser.AI is designing the AI of the future, an AI that will augment workers, create jobs, and elevate humanity. If you are interested in making a real SEC Form C investment in the future, please visit us at WeFunder (link in comments)

Comment on visual. Even though I went to Harvard I was always pissed off that I didn't go to MIT because they have the coolest Beaver Ring.  I mean that ring is too cool for school. Who doesn't want a Beaver ring?

BREAKING: The $400 Billion U.S. AI Infrastructure Buildout May Have Just Become a Stranded Asset

It may not be a lack of capital that ends AI 1.0 and the era of the behemoth LLM, it may simply be smarter technology.

Better design beats brute force any day of the week.

See there's this 23-page paper about architecture.

DeepSeek just published research describing manifold-constrained hyper-connections (mHC), an approach that dramatically improves training stability without brute-force scale.¹

The numbers:²

•	Signal amplification during training reduced from 3000× to stable
•	+2.1% improvement on reasoning benchmarks (BBH) for 27B models
•	Achieved on 3B, 9B, and 27B parameter models
•	With roughly 6-7% compute overhead

Read that last number again.

**6-7% overhead to solve a problem the entire industry has been throwing hundreds of billions at**

For three years, the dominant AI narrative has been simple: bigger models = more intelligence. 

That story justified $342 billion in hyperscaler capex for 2025.³ A $500 billion Stargate project.⁴ Data center construction hitting $40 billion in a single month.⁵

DeepSeek asked a different question: 

What if the instability everyone compensates for with brute force could be removed through better design?

They showed it could.
This is the kind of innovation that almost never comes from incumbents. Not because they lack talent. Because they lack freedom.

The major U.S. labs aren't just technology companies anymore. 
They're capital structures. 
Their architectures are coupled to multi-year infrastructure bets and investor narratives that assume scale equals progress.

A true architectural reset, one where smaller, more efficient models match or beat bloated ones—doesn't just threaten product roadmaps.

It threatens the entire financial logic underneath them.

So the incentives point the wrong direction. The people who most need to pivot are the people least able to.

We've seen this movie before:
Toyota didn't beat Detroit by building bigger factories.
ARM didn't beat Intel by designing hotter chips.
SpaceX didn't beat Boeing by spending more per launch.
Constraint forces clarity. Capital abundance becomes its own trap.

I'm not saying DeepSeek's specific architecture is the final answer. But they've demonstrated something the industry has been quietly dreading:
The frontier isn't just about scale anymore.
It's about intelligence per dollar. Per watt. Per parameter.

That metric favors the lean over the lavish. 
The constrained over the capitalized. The teams who have to think over the teams who can just spend.
The next few years in AI won't be won by whoever has the most GPUs.

They'll be won by whoever understands what intelligence actually costs, and builds accordingly. (sources in comments)
-
“The trick with technology is to avoid spreading darkness at the speed of light.”
Invest in AI 2.0 and own a piece of what comes next. We are raising on WeFunder (Link in comments)

Apple won’t win the AI race.
They’ll redefine it.

(Disclosure: I am a huge Apple fan, ever since Steve Jobs, and I love their strategic silence and grown-up presence in an industry filled with rampant immaturity and juvenile hysteria. I miss Steve Jobs)

For the last year, the narrative has been:
Apple is behind in AI.

I think that misunderstands Apple entirely.

Apple doesn’t ship technology because it’s impressive.
They ship when it’s civilized.

GenAI 1.0 is powerful, but it’s also noisy, probabilistic, hallucination-prone, and optimized for speed over judgment. 

And no one is actually paying for it, including the enterprise, who stopped buying it for all intents and purposes.

It's unreliable and sloppy.
It's good enough.

That’s acceptable for chatbots and copilots. It’s unacceptable for an operating system that sits at the center of identity, health, payments, and trust.

Apple doesn’t build systems that “usually get it right.”
They build systems that know when not to speak.

So instead of a Copilot clone or a chatty assistant, Apple’s likely move is quieter and more dangerous to incumbents:

• AI as infrastructure, not interface
• Many small, task-specific models instead of one giant brain
• On-device first, bounded behavior, deterministic fallbacks
• Fewer words, fewer guesses, fewer mistakes

The AI becomes invisible and becomes useful behind the scenes.

In other words: intelligence as judgment, not generation.

This is also why Siri won’t become louder or more “Agentic.”
It will dissolve into the OS, less visible, more disciplined, more opinionated about when to stay silent.

The irony is that when Apple finally ships its version of AI, many people will say:

“That’s it?”

And that will be the tell.

Because Apple has never tried to win hype cycles.
They wait for the noise to peak, then redefine the finish line.

If Steve Jobs really wanted “Aristotle in every pocket,” it was never about answers on demand.

It was about discernment.

Apple isn’t late.
They’re refusing to ship intelligence that is unintelligent
********************************************************************************
"THE TRICK WITH TECHNOLOGY IS TO AVOID SPREADING DARKNESS AT THE SPEED OF LIGHT"
Curiouser.AI is building AI 2.0, backed by a community who believe we can design AI to elevate humanity. If this resonates please try our beta at curiouser.ai or visit WeFunder (Link in comments)

Major Tech CEOs Are Scrambling In Real Time to Reposition Themselves
They will now do an about face to try and do damage control

They'll look like the scarecrow in the Wizard of Oz pointing here and there and everywhere

IBM and Google can hedge.
OpenAI and NIVIDIA can't.

From “we can’t invest enough in AI”
to
“the industry may be overinvesting.”

Watch closely. This is the largest corporate repositioning exercise since executives claimed they “always believed in mobile.”

Last week, IBM CEO Arvind Krishna said on The Decoder podcast:

“There’s no way you’re going to get a return on that.
$8 trillion of capex means you need roughly $800 billion of profit just to service the interest.”¹

Then Google CEO Sundar Pichai told the BBC that today’s AI spending shows “elements of irrationality” adding:

“I don’t think any company is immune. Including us.”²

Ah yes. Now they tell us.

Because here’s what Pichai said just months earlier, in July 2024:

“The risk of underinvesting is dramatically greater than the risk of overinvesting.”³

Zuckerberg said the same thing.
Microsoft echoed it.
It was the synchronized talking point of the moment.

Now the tone has shifted.

Krishna went further, quietly noting that today’s AI has a 0–1% chance of reaching AGI.⁴

Not 50–50.
Not “uncertain.”
Zero to one percent.

Here’s the pattern:

Phase 1: We must invest at all costs or be left behind.

Phase 2: There may be some irrationality here.

Phase 3 (coming soon): We were appropriately cautious all along.

This isn’t analysis.
It’s liability management.

When the write-downs arrive, and with ~$8 trillion in committed infrastructure spend, they will, no CEO wants to be on record cheerleading the week before.

The tell? Watch who can’t hedge yet.

OpenAI can’t, their valuation depends on the narrative.
Nvidia can’t, Jensen Huang has dismissed bubble concerns outright.⁵

The companies that can survive a correction are already softening the ground.
The ones that can’t are still pumping.

Krishna’s math is brutally simple:

A 1-gigawatt data center costs ~$80B to build

Commit to 20–30 GW and you’re staring at $1.5T in capex

Chips depreciate in ~5 years full replacement required

That’s not a business model.
That’s a treadmill set to bankruptcy.

Yes, the enterprise AI market will be massive.
Krishna himself estimates trillions in productivity gains.

But there’s a canyon between:

AI is useful
and

$8 trillion in infrastructure to chase AGI makes financial sense

The first is true.
The second is what’s now being quietly walked back.

The repositioning has begun.
********************************************************************************
“The trick with technology is to avoid spreading darkness at the speed of light.”
Curiouser.AI is backed by a community of thoughtful people all of whom believe we can design AI to augment work, create jobs, and elevate humanity. If this resonates please try our beta at curiouser.ai or visit WeFunder (Link in comments)

How Many Tech Companies Do You Trust? Make a list.
(I can think of one I really trust)

Now ask yourself: how many tech companies are competing for your attention?
That gap, infinite attention-seekers, near-zero trust-holders, is the entire market failure of AI 1.0 in one frame.

We've been living in the Attention Economy. The winners captured eyeballs by whatever means necessary.

But attention is now worthless. AI can manufacture infinite content at near-zero cost. 

The AI content creation market is growing at 32.5% annually.¹ 83% of marketers say AI helps them produce "much more content."² AI-generated articles briefly surpassed human-written articles in late 2024.³

When everything competes for attention, attention stops being strategic currency.

What can't be manufactured becomes scarce. And trust, by definition, cannot be manufactured. Only earned.

Welcome to the Trust Economy.

The data is devastating: A decade ago, 73% of Americans trusted technology companies. Today? 63%.⁴ 

Trust in AI companies specifically has declined from 62% to 54% over five years.⁵ 

Only 32% of Americans trust AI.⁶

Meanwhile, 85% of customers now hesitate to engage with a company if they have doubts about its data practices.⁷

Here's what makes this profound: the traditional moats are eroding simultaneously.

IP? AI is raising the "obviousness" bar for patents. If an AI-enhanced person skilled in the art can arrive at your invention, it's not patentable.⁸ 

Your proprietary model gets reverse-engineered before your lawyers finish filing.

Regulation? The referees are leaving the field. Consumers must protect themselves, which means choosing who to trust.

Automation? 95% of enterprise AI pilots fail to deliver measurable business impact.⁹ 42% of companies abandoned their AI initiatives in 2025, up from 17% in 2024.¹⁰ 

When everyone automates, automation is table stakes.

Strip away IP protection, regulatory guardrails, and technical differentiation. 

What remains?

The relationship.
The irony: the very forces that were supposed to make business more efficient

AI, deregulation, automation, are forcing a return to something ancient. Reputation. Your word. The handshake.

Except now it operates at scale.

The AI 1.0 players optimized for everything except the thing that's about to matter most.
-
“The trick with technology is to avoid spreading darkness at the speed of light.”
Invest in AI 2.0 and own a piece of what comes next. We are raising on WeFunder (Link in comments)

Sources:
¹ Grand View Research, Generative AI Content Creation Market 2024
² SQ Magazine, AI in Social Media Tools Statistics 2025
³ Axios/Graphite analysis of 65,000 URLs, Oct 2025
⁴ ⁵ ⁶ 2025 Edelman Trust Barometer
⁷ PwC 2025 Global Digital Trust Insights
⁸ NovoTech Patent Firm analysis, AliveCor v. Apple, March 2025
⁹ MIT NANDA Project, "The GenAI Divide," 2025
¹⁰ S&P Global Market Intelligence 2025 Enterprise Survey

taplio

The Battle

Chris M.

Stephen Klein

Followers

Posts

Engagements

Likes

Performance ScoreStephen Klein's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Chris M.

Stephen Klein

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Stephen Klein's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreStephen Klein's LinkedIn Performance