Stephen Klein's Linkedin Analytics

(Please note: I fell down the rabbit hole. I saw how much money the Gen AI companies were losing and asked: Then who is making all the money?) Management Consulting Firms Are Monetizing Fear at Scale The average CAGR for the top 5 management consulting firms in Generative AI year over year for 2021-2025 105.9% While startups bleed cash and foundational model companies lose billions on every product, five consulting firms: Accenture, BCG, Deloitte, McKinsey, and PwC, have quietly built the fastest-growing revenue engine in the GenAI ecosystem. In fact they are the ONLY ones making money. Accenture grew its GenAI revenue from $100M in 2021 to over $2B in 2024. It booked $1.2B more in just Q1 of 2025.1 BCG, Deloitte, McKinsey, and PwC are following suit, doubling revenue year over year, spinning up GenAI advisory units, and securing massive enterprise transformation deals.2345 Every major firm now has a GenAI services practice. All of them are billing millions. What Are They Selling? These firms don’t build models. They don’t run GPUs. They don’t even need to prove GenAI works. They sell: Strategy frameworks Integration blueprints Responsible AI protocols Change management programs Confidence in chaos And the Clients? How are They Doing? We don't really know. The “success stories” we hear about GenAI come from: Reports authored by the same consulting firms selling the services Surveys funded by GenAI vendors Case studies with no external verification “Generative AI could add $4.4 trillion annually to the global economy.” McKinsey, 20234 Or this: “85% of leaders say GenAI will improve competitiveness.” BCG, 20242 But where is the proof that companies are actually profiting? Where are the independent audits? The financial statements showing actual ROI? The success metrics that didn’t originate from the people getting paid to say it’s working? Strategic Takeaway Model builders? Losing billions. Tool vendors? Burning capital. Enterprises? Still figuring it out. Consultants printing money, growing their GenAI revenue over 100% per year. Before you commit millions, ask: Who produced this data? Who profits from this narrative? Who is actually making money? In GenAI: consultants are monetizing uncertainty at scale. ******************************************************************************** The trick with technology is to avoid spreading darkness at the speed of light Stephen Klein is Founder & CEO of Curiouser.AI, the world’s first Reflective AI designed to augment, not replace, human intelligence. He also teaches AI ethics and strategy at UC Berkeley. To signup visit curiouser.ai or contact hubble https://lnkd.in/gphSPv_e Footnotes Accenture passes $2B in GenAI revenue – FNLondon ↩ BCG GenAI transformation report – BCG.com ↩ ↩2 Deloitte FY2024 Global Report ↩ McKinsey: The economic potential of GenAI ↩ ↩2 PwC commits $1B to GenAI – PwC Global ↩

You Are Becoming Average Right now, tens of thousands of professionals are leaning into prompt engineering like it’s the new literacy. "If I could only prompt better I would be so much more successful!" "Thank God there are all these experts ready to show me how!" They’re refining, tweaking, and even buying “elite” prompts. But beneath that surface effort is a statistical reality no one’s talking about: The more you prompt, the more you regress to the mean. You prompt GPT, Claude, Gemini, or Mistral: “Give me 5 SaaS trends for 2025.” “Write a bold B2B landing page.” “Create a strategy deck intro slide.” But here’s what you probably haven’t considered: Thousands of other users typed similar prompts. You’re all working with the same model. You’re all pulling from the same probability-weighted token distributions. Enter: Regression to the Mean Regression to the mean is a statistical phenomenon where extreme or standout outcomes tend to move closer to the average over time In GenAI: That “amazing” result you got? It’s partially luck. When you reuse the prompt, or others copy it? The luck disappears. The model returns to its mean behavior, patterned, safe, and undifferentiated. Viral prompts decay. Shared prompts homogenize. And the more people prompt, the less anyone stands out. Prompting feels like personalization, but it’s actually mass production. Prompt fatigue: Diminishing returns from the same tricks Prompt collapse: Model updates break prompt behavior Output homogenization: Everyone converges on the same tone, language, and ideas What Smart Builders Are Doing Instead Designing model-agnostic architecture Creating original IP, not just output Building RAG + orchestration layers that don’t rely on prompting Fine-tuning local or open-source models for internal context Focusing on system-level thinking, not surface-level prompting If your strategy is to “get really good at prompting,” your strategy is to regress to the mean. And the mean is getting lower every day. Prompting isn’t bad. But it likely means you're becoming more average than you think ******************************************************************************** The trick with technology is to avoid spreading darkness at the speed of light Stephen Klein is Founder and CEO of Curiouser.AI, the only Generative AI designed to augment human intelligence. He teaches AI Ethics at UC Berkeley. To signup visit curiouser.ai or connect on hubble https://lnkd.in/gphSPv_e Sources [1] Regression to the Mean | Scribbr. https://lnkd.in/guC67e-a [2] Farnam Street. Regression Toward the Mean: A Mental Model. https://lnkd.in/grcYAxUC [3] Stanford HAI. (2024). Foundation Model Transparency Index. https://hai.stanford.edu [4] Hugging Face. (2024). Open LLM Leaderboard. https://lnkd.in/g3cX-Mwx [5] Arxiv.org. (2023). The False Promise of AI Productivity Metrics. https://lnkd.in/gjBuJ3Tk

Science Has Figured Out What Businesses Can't They're the grownups in the GenAI world, while business sits at the kiddie table The scientific community has been remarkably clear on what GenAI is, and what it isn’t, for years. They are solving problems and using GenAI as the ultimate augmentation partner. It's breathtaking to see what they've accomplished. But business leaders continue to put on a theatrical performance with their management consulting producers Here’s what science has known for a while: It doesn’t reason. It predicts. It can't replace anyone, it's not qualified. It doesn’t understand. It correlates. It needs to be managed as it can’t be trusted It doesn’t “know” truth. It reflects training data. It doesn’t have intent, memory, or values. The confusion in business surrounding GenAI leads to: Misguided Investments and Lazy Analogies Companies imagining GenAI as a “smart intern” waste money on unrealistic expectations. Failed Pilots When treated like a “Chief Productivity Officer,” GenAI is tested with human-like productivity assumptions, resulting in failure rates as high as 70%.⁵ Inflated Expectations The hype about “automating everything” leads to disillusionment when GenAI doesn’t deliver on promises. There is no Hype-as-a-Service (HaaS) in science. In science, GenAI is used with clear, grounded expectations, augmentation, not magic. In business, HaaS rules, where the focus is more on selling the promise of AI, not the reality. Instead of chasing efficiency at all costs, we need to ask the right questions: How can GenAI elevate human judgment? How can it augment creativity, not automate routine tasks? How can it help us see patterns we couldn’t see before? It’s a mirror, not a mind. The Path Forward: First Principles, Not Metaphors Think about the core capabilities of GenAI and how it fits into your business. Clear Thinking, Not Brand Buzzwords Use clear, accurate language to understand GenAI’s real potential, and avoid marketing speak that clouds judgment. Augmentation, Not Automation Embrace GenAI as a tool to enhance and elevate human capabilities, not replace them. ******************************************************************************** The trick with technology is to avoid spreading darkness at the speed of light Stephen Klein is Founder & CEO of Curiouser.AI, the world’s first values-based AI platform, strategic coach, and advisory. He also teaches AI strategy and ethics at UC Berkeley. To learn more visit curiouser.ai or connect on hubble at https://lnkd.in/gphSPv_e Footnotes: Radford et al., GPT: Improving Language Understanding by Generative Pre-Training, OpenAI, 2018 Bender et al., On the Dangers of Stochastic Parrots, FAccT, 2021 Weidinger et al., Taxonomy of Risks Posed by Language Models, DeepMind, 2022 Mitchell, M., Artificial Intelligence: A Guide for Thinking Humans, 2021 McKinsey, The State of AI in 2024, 2024

Sam Altman is About To Turn Into A Pumpkin And that's why he's back in Middle East Sam Altman introduced Generative AI to the world. (And for that we owe him a highly qualified debt of gratitude) But now he has 5-6 months of life left. Despite raising $40B in a headline grabbing round, OpenAI is burning so much cash that it's likely it only has just 5–6 months of runway left. Why? Because he is likely to only receive $10B of the $40B-- $30B of it was contingent on OpenAI becoming a for-profit company. That transition died last week. Meanwhile: Estimated annual burn: ~$7B Monthly burn: $600M–$1B Product margin: negative, even the $200/month Pro subscriptions loses money (Can't make that up in volume people, the more you sell, the more you lose) GPT-4 Turbo & enterprise deals? Underpriced, subsidized, and unsustainable Error rates steadily getting worse (this alone is a deal breaker!) DeepSeek can match his performance at a 98% discount (Yea they stole the data but they stole the data he stole first) 4.5 was an embarrassment. Operator did little more than lend some energy to the agentic hype. His imaging tool that he disrespected Miyazaki and the Japanese people with, an expensive toy. This is not a company. This is a cash burning hype machine like the world has never seen. As far as I can see, Altman has three possible lifelines, none of which are rooted in sound business fundamentals and never were: 1. A Middle East Bailout He’s courting sovereign partners to build AI infrastructure, cheap water, cheap energy, and cash-rich stability. But geopolitics isn’t a business model. (But it's a worry for America) 2. A U.S. Government Bailout Altman is positioning OpenAI as essential to American competitiveness against China. It’s a clever narrative, but it's also a bet on federal subsidies, not market viability, though he is now close to Trump) (He has wrapped himself in the American flag. DeepSeek may be the best thing to have ever happened to him). 3. Another Silicon Valley Ante-Up VC's and tech giants may come back to the table, not because it’s a good bet, but because they’ve already invested too much. They could and likely will lose it all in a collapse (unless they've been quietly monetizing cash on the inside). And all of this is happening while OpenAI faces: A lawsuit from Elon Musk over mission betrayal Delays in its $100B data center project due to tariffs Microsoft hedging No competitive moat. You can’t burn $1 billion a month forever. Or can you? ******************************************************************************** The trick with technology is to avoid spreading darkness at the speed of light Stephen Klein is Founder & CEO of Curiouser.AI, a values-based AI platform and advisory helping leaders build sustainable, augmentation-first strategies. He also teaches AI Ethics at UC Berkeley. Learn more at curiouser.ai or connect on hubble https://lnkd.in/gphSPv_e