Thoughts on AI research, machine learning, safety, and technology
A comprehensive survey maps the shift from fast, intuitive System 1 AI to deliberate, step-by-step System 2 reasoning — covering everything from Monte Carlo Tree Search to reinforcement fine-tuning, and what it means for the next generation of AI agents.
Indiana University researchers used simple prompts to extract near-complete copyrighted text from Claude, GPT-4o, and Llama — revealing that memorization is a fundamental byproduct of next-token prediction, not a solvable bug.
AI benchmarks drive model development. So why are we still evaluating humans with whiteboards and trivia? It's time to rethink hiring around human-AI collaboration and real deliverables.
A deep dive into catastrophic forgetting and modern approaches to lifelong learning in AI systems.
How modern AI systems learn to understand and connect visual and textual information.
Exploring how LLMs are transforming the way we approach scientific discovery and research methodology.
From the Great Depression to trade wars to rising nationalism, the pattern repeats every century. The root cause isn't economics or politics — it's that the majority of people never learned why it happened the first time.