The hidden ceiling in AI

The hidden ceiling in AI

What’s up, everyone – Pranjal here. Welcome back to Generative Finance, the newsletter on AI x fintech.


Let’s jump straight in this week.



?

My favorite finds of the week.

  • The AI tipping point (link)
  • Could making money from AI be a challenge for banks? (link)
  • The DORA effect (link)
  • Fintech business weekly (link)




NEWS

The delicate dance of human and digital intelligence

The backstory: As AI sweeps through finance, a fascinating tension is emerging. At Tearsheet's Power of Payments conference, Silicon Valley Bank's Head of Product set the tone: while everyone's rushing to digitize everything, there's a risk of "falling off a cliff if you need to talk to a human being who knows anything beyond what the systems are doing."

Now... Key players are drawing lines in the sand:

  1. SVB is deliberately maintaining human expertise alongside AI capabilities
  2. Payoneer is using AI to remove barriers in global commerce
  3. ZIP Co is deploying AI to understand spending patterns across multiple shopping environments
  4. Fifth Third warns about prioritizing growth over stability in fintech

THE TAKEAWAY While fintech startups optimize for speed and efficiency, traditional banks are grappling with a harder question: how to preserve the human judgment that's kept banking stable for centuries. Consider this: When SVB talks about not "over-rotating" to digital, they're not being luddites - they're acknowledging that banking is fundamentally about trust, not just transactions. For finance leaders watching closely: the real innovation challenge isn't teaching AI to handle payments faster, it's figuring out how to blend human expertise with AI capabilities without losing what makes banking work in the first place. In an industry rushing to digitize everything, maintaining the human element is a challenge work thinking about.




MY TAKE

The hidden ceiling in AI (and why it might not matter)

We've reached a pivotal moment in artificial intelligence. The current generation of AI models have already ingested virtually all publicly available human knowledge - billions of pages of text, academic papers, books, websites, and digital archives. We've essentially fed these systems the entire internet and every publicly accessible document humanity has created.

And yet, despite having consumed more information than any human could read in multiple lifetimes, these models have hit a curious plateau. They can answer questions, generate text, and perform tasks with remarkable breadth, but they struggle to demonstrate the kind of nuanced, adaptive intelligence that defines professional expertise.

The challenge isn't about knowledge accumulation anymore. These models know almost everything that's been written down. The limitation is experiential learning - the tacit knowledge that can't be captured in text.

Professional expertise is fundamentally about skills that aren't documented: how a lawyer intuitively understands when to settle a case, how a therapist skillfully navigates complex emotional landscapes, how a salesperson builds genuine connection and trust. These are skills learned through doing, through failure, through subtle interactions that don't get written into instruction manuals.

As Ilya Sutskever and other AI researchers have noted, we've essentially exhausted our existing data sources. Adding more text won't significantly improve model capabilities. The next breakthrough won't come from more information, but from new learning methodologies.

Two promising paths are emerging:

First, reinforcement learning offers a potential breakthrough. This is where AI has already shown remarkable capabilities - in game-playing, strategic decision-making, and emerging fields like self-driving technology. By creating environments where AI can learn through trial and error, we might develop more adaptive intelligent systems.

The second approach involves breaking professional tasks into smaller, independently solvable components. This method involves creating AI systems that can coordinate complex tasks by solving individual elements and then integrating them intelligently.

The future of AI isn't about creating an all-knowing system, but about developing sophisticated tools that can learn, adapt, and collaborate with human professionals in increasingly nuanced ways.

Rather than a dead end, this is an exciting frontier. Each challenge in AI development is an opportunity to reimagine how technology can enhance human capabilities. The next few years will likely see AI becoming more specialized, more context-aware, and better at understanding the complex, often unspoken dynamics of professional work.

The most profound technological innovations happen at the intersection of what we know and what we're just beginning to imagine. AI isn't replacing human expertise - it's expanding our collective potential, offering new ways of solving complex problems and understanding the world around us.

Until next time,

Pranjal



?

How I can help

We can help speed up your compliance and onboarding process.

We built Accend as your AI-powered platform to help risk and compliance teams get customers onboard quicker. Get started today with us today.

要查看或添加评论,请登录

Pranjal Daga的更多文章

  • FDIC's Secret List & Hyper-Evolution

    FDIC's Secret List & Hyper-Evolution

    What’s up, everyone – Pranjal here. Welcome back to Generative Finance, the newsletter on AI x fintech.

  • Marc Andreessen, Joe Rogan, and the CFPB

    Marc Andreessen, Joe Rogan, and the CFPB

    What’s up, everyone – Pranjal here. Welcome back to Generative Finance, the newsletter on AI x fintech.

    9 条评论
  • Revolut's Roadmap + Old Fraud Data

    Revolut's Roadmap + Old Fraud Data

    What’s up, everyone – Pranjal here. It’s almost Thanksgiving.

  • Klarna's IPO, Stripe's AI SDK + WeChat of the West

    Klarna's IPO, Stripe's AI SDK + WeChat of the West

    What’s up, everyone – Pranjal here. Welcome back to Generative Finance, where we talk all things fintech x AI.

  • Deceptive Tips + An Identity Crisis

    Deceptive Tips + An Identity Crisis

    What’s up, everyone – Pranjal here. In today’s edition: Dave gets busted over manipulative tip strategy Fintech…

    5 条评论
  • Bad Metrics, Peak Regulation + Credit Suisse Meltdown

    Bad Metrics, Peak Regulation + Credit Suisse Meltdown

    What’s up, everyone – Pranjal here. After a packed - but great - few days at Money2020, I'm glad to be back in New York.

    2 条评论
  • Apple, CFPB + Excel Spreadsheets

    Apple, CFPB + Excel Spreadsheets

    What’s up, everyone – Pranjal here. There was a LOT that happened this week.

    2 条评论
  • Axiom, Whistleblowers + Good Data

    Axiom, Whistleblowers + Good Data

    What’s up, everyone – Pranjal here. Another week, another compliance scandal ?? Let’s jump straight in! My favorite…

    1 条评论
  • Scams, TD Bank + Losing the Arms Race

    Scams, TD Bank + Losing the Arms Race

    What’s up, everyone – Pranjal here. We’re back with another edition of Generative Finance - the best in finance x AI…

    2 条评论
  • Dynasties, Stripe + Genuine Delight

    Dynasties, Stripe + Genuine Delight

    What’s up, everyone – Pranjal here. Welcome to the first issue of Generative Finance – where we dive headfirst into the…

    15 条评论

社区洞察

其他会员也浏览了