Dwarkesh Patel hosts MatX CEO Reiner Pope for a blackboard lecture tracing chip design from basic logic gates through GPUs, TPUs, FPGAs, and the computational structure of the human brain.
Dwarkesh Patel
dwarkesh.com · Leading Thinkers · 12 items
Dwarkesh Patel argues that intelligence and power are distinct capacities, challenging assumptions underlying many AI risk arguments and AGI forecasts.
Dwarkesh Patel shares notes on LLM pretraining parallelism strategies and the often underappreciated causes of large-scale training run failures.
Dwarkesh Patel argues that reinforcement learning from verifiable rewards (RLVR) may be systematically disadvantaged in scientific domains where correct answers cannot be easily checked.
Dwarkesh Patel hosts Eric Jang for a walkthrough of building AlphaGo from scratch using modern AI tools, using the exercise to illuminate the primitives of current AI systems.
Dwarkesh Patel hosts geneticist David Reich to discuss his new paper (with Ali Akbari) overturning the consensus that natural selection has been dormant in humans since the Bronze Age.
Dwarkesh Patel hosts Reiner Pope for a blackboard lecture on the mathematical foundations of how frontier LLMs are trained and served at scale.
Dwarkesh Patel extends his list of big open questions about AI and offers a blog prize for compelling answers, as part of a search for an intellectual co-researcher.
Dwarkesh Patel announces a blog prize for the best responses to his big open questions about AI, seeking high-quality intellectual output on the most pressing issues.
Dwarkesh Patel interviews Nvidia CEO Jensen Huang on competing with Google TPUs, the debate over selling AI chips to China, and Nvidia's manufacturing supply chain advantages.
Dwarkesh shares weekly AI learnings covering whether distillation can be restricted, the cybersecurity equilibrium following Claude Mythos, and advances in Pipeline RL training.
Dwarkesh Patel interviews Michael Nielsen on how scientific progress actually happens, with implications for closing the RL verification loop for AI-driven discovery.
Nothing matches.