All Videos

Stanford Online

Fine-tuning large language models risks sudden catastrophic failure of safety guardrails, which break abruptly rather than gradually like capability metrics. Researchers demonstrate that dynamically segmenting training data into safe and unsafe portions—instead of binary filtering—maintains both safety alignment and model performance.

a16z Podcast

AI infrastructure deployment is unprecedented with $400B in annual capex from tech giants, while input costs have plummeted 99% and user adoption occurs 5x faster than previous tech cycles. However, energy availability and cooling technology will become the critical bottlenecks constraining growth over the next five years, even as business models evolve to capture value through price discrimination and high retention rates.

Corey Schafer

This tutorial demonstrates how to connect JavaScript frontend forms to a FastAPI backend using the fetch API, implementing full CRUD functionality through Bootstrap modals while temporarily hardcoding user authentication and centralizing business logic like data sorting in SQLAlchemy queries.

Huberman Lab

Dr. David Eagleman explains that neuroplasticity allows the human brain to constantly reconfigure its 86 billion neurons throughout life, emphasizing that maintaining this flexibility requires continuously seeking novel challenges rather than repeating mastered skills, while modern tools like AI can enhance critical thinking when leveraged through curiosity-driven inquiry.

CNBC

Claudebot represents a significant leap toward fully autonomous AI agents by combining proactive task generation with open-source flexibility, but its lack of an ethical 'aspirational layer' highlights the urgent need for safety frameworks like the Heuristic Imperatives to guide these systems.

Conversations with Tyler (Tyler Cowen)

Tyler Muir explains that Quantitative Easing operates primarily as a state-contingent commitment that backstops financial intermediary balance sheets during crises, challenging traditional "Wallace neutrality" by demonstrating that asset prices are set by constrained intermediaries rather than representative households, creating immediate market impacts through policy announcements alone.

Cognitive Revolution

Dominic Williams outlines the Internet Computer as a 'sovereign cloud' enabling tamperproof, unstoppable applications built by AI through natural language prompts, representing a fundamental reimagining of cloud computing that intersects with critical debates around AI safety and decentralized control.

Machine Learning Street Talk

Dr. Jeff Beck argues that agency cannot be verified from external behavior alone, requiring instead evidence of internal planning and counterfactual reasoning, while advocating for energy-based models and joint embedding architectures as biologically plausible alternatives to standard function approximation.

We Study Billionaires (TIP)

Despite Meta trading at a significant discount to tech peers (22x forward PE) and improving its core ad business with AI, the market is pricing in skepticism about $100B annual AI investments. The Reality Labs division has burned $70B with limited success in VR, and Meta faces steep competition from Google's Android XR ecosystem in the AR glasses race, raising questions about whether this massive capex will generate returns or simply dilute the core business.

The Pomp Podcast

The video argues that silver is being repriced as a critical technology mineral for AI and drones, triggering a capital rotation from gold to silver that will soon mirror in crypto with Ethereum leading Bitcoin higher. A massive short squeeze in Bitcoin is expected this year as the two-year whale redistribution phase ends, making current prices the final opportunity to enter before a parabolic breakout above $100,000.

Ben Felix

Despite headline GDP growth hitting 4.4% in Q3 2025, the U.S. economy is experiencing a severe bifurcation where asset owners and AI-driven investment thrive while the labor market contracts and tariffs force middle-class consumers into expensive domestic substitutions, masking true economic health for the bottom 90%.

Stanford Online

This seminar outlines research on resilient autonomy for robots in degraded environments like underground mines and caves, emphasizing unified perception models that process pixels once for multiple geometric tasks, multi-modal sensor fusion (thermal, IMU, radar), and novel representations enabling long-range semantic reasoning without depth.