A discussion between cognitive scientist Gary Marcus and physicist Brian Greene on the true capabilities and structural limitations of current generative AI models. Marcus argues that while large language models are proficient at mimicking the surface-level appearance of logical reasoning, they lack the underlying cognitive architecture required to generalize beyond their training sets.
Guessing with Style
AI is an incredible force multiplier for engineers who already know how to build a system. But if we outsource architectural intuition to a transformer model without grounding it in raw telemetry and end-to-end load testing, we aren't engineering anymore—we're just guessing with style.
Yann LeCun’s lecture at ETH Zürich outlining his vision for the next generation of artificial intelligence. He contrasts current autoregressive LLMs with energy-based world models capable of reasoning, planning, and understanding causality, arguing that true machine intelligence requires learning from passive observation rather than simple next-token predictions.
Prototyping Myopia
It is a case of hyper-fixated technological optimism leading to algorithmic deference. The leader, seduced by the conversational fluency and immediate output of the tool, undergoes a form of prototyping myopia—over-indexing on isolated AI capabilities while losing their strategic anchor in the broader operational reality of the business.
A critical and clear-eyed look at the capabilities and limitations of modern artificial intelligence. The authors dismantle the hype surrounding machine learning, showing where it succeeds, where it fails, and why predictive AI often amounts to little more than snake oil. It serves as an essential guide to distinguishing genuine technological progress from empty marketing claims.
Grant Sanderson’s brilliant visual introduction to the mathematics of deep learning. Using intuitive animations, the series breaks down what a neural network actually is, how gradient descent works, and the core calculus behind backpropagation.
Andrej Karpathy’s masterclass series on building neural networks from scratch in Python. Starting from basic backpropagation up to building a full GPT-style transformer, this playlist is widely considered one of the most accessible and thorough resources for understanding deep learning from first principles.
AI pioneer Stuart Russell’s compelling analysis of the long-term risks posed by superhuman artificial intelligence. The book dismantles standard arguments downplaying AI risk and proposes a new foundation for AI development centered on provably beneficial machines. Russell argues that we must rebuild AI around the core principle of machines that are uncertain about human preferences, ensuring they always remain subservient to humanity.