The Lifelong Study: Insights From Rotterdam11 Maalis

Why general artificial intelligence is a myth?

Welcome back to the podcast! Today we are diving into a thought-provoking new paper titled 'AI Must Embrace Specialization via Superhuman Adaptable Intelligence' by researchers including Judah Goldfeder and Yann LeCun.

If you follow the tech world, you hear the term AGI—Artificial General Intelligence—thrown around constantly by executives, doomsayers, and politicians. But this paper argues that our entire obsession with AGI is fundamentally flawed because it relies on the illusion that human intelligence is 'general'.

The authors argue that humans aren't actually general intelligence engines at all; we are highly specialized creatures who were honed by evolution to master a very narrow set of skills required for survival.Instead of chasing the myth of AGI, the authors propose a new North Star for the field: Superhuman Adaptable Intelligence, or SAI.

Rather than trying to build a single, monolithic AI that can do literally everything, SAI focuses on adaptability—specifically, the speed at which an AI can learn and master new, important tasks.

As the authors brilliantly put it: 'The AI that folds our proteins should not be the AI that folds our laundry!'.

This paper makes a compelling case that to push the boundaries of technology, we need to abandon our human-centric biases and lean into specialized models powered by self-supervised learning and world models. Let's get into it!"

Tämä jakso on lisätty Podme-palveluun avoimen RSS-syötteen kautta eikä se ole Podmen omaa tuotantoa. Siksi jakso saattaa sisältää mainontaa.

Jaksot(28)

How Much Data Is Enough? Why the Riemann Zeta Function Predicts the Future of AI & Discovery

In biomedical AI, we constantly ask: Do we need more data, or do we need a smarter model?In this episode, we break down The Zeta Law of Discoverability—a theoretical framework linking sample size comp...

30 Elo 34min

Taps, Leaks, and Plugs: Explaining the Worldwide Rise in Parkinson's

Why are global Parkinson's disease cases projected to skyrocket to 25.2 million by 2050?In this episode, we explore an ingenious "bathtub analogy" to explain this massive growth. We break down the "in...

6 Maalis 19min

Hallmarks of Cancer: The Paper That Defined a Field

Few scientific papers truly define a field. The Hallmarks of Cancer is one of them.In this episode, I dive into the evolution of the Hallmarks framework—where it started, how it has shaped decades of ...

1 Helmi 18min

Closing the Women’s Health Gap: From History to Future Innovation

This episode dives into the history and persistent gaps in women’s health research — from decades of exclusion in clinical trials to today’s underfunding of critical conditions like menopause, autoimm...

16 Syys 202519min

Beyond the Buzz: Unmasking Bias in Healthcare AI, From Conception to Clinic

Our latest podcast episode delves into the critical paper: "Bias recognition and mitigation strategies in artificial intelligence healthcare applications" . This essential review highlights how AI, de...

2 Syys 202524min

Rotterdam Studie Bijgewerkt 2024

De Rotterdam Studie is een langlopende, populatie-gebaseerde cohortstudie in Nederland die zich richt op het identificeren van risicofactoren en het begrijpen van het verloop van multifactoriële aando...

1 Touko 20257min

Beyond Human Data: How AI Will Learn Like We Do

We’re standing at a turning point in AI. Until now, machines have learned by studying us — reading our books, analyzing our conversations, mimicking our decisions.But what happens when AI starts learn...

28 Huhti 202514min