@Nona_xai: Google DeepMind chip engineer Reiner Pope just explained on a whiteboard what no one had ever explained to you before: …
Summary
Google DeepMind chip engineer Reiner Pope delivers a comprehensive whiteboard explanation of how chips work, covering logic gates to systolic arrays and the human brain, in a free YouTube video.
View Cached Full Text
Cached at: 05/23/26, 10:17 PM
Google DeepMind chip engineer Reiner Pope just explained on a whiteboard what no one had ever explained to you before: how a chip really works, from the logic gate to the human brain.
In 1h15 he covers everything that engineering schools take years to teach.
→ How to build multiplication from scratch. → Why moving data costs more than computing. → How systolic arrays work that make LLMs run. → Why a CPU core is 10 times larger than a GPU one. → Why a GPU is nothing more than a bunch of small TPUs. → And why the human brain remains an enigma for hardware architects.
It’s free. It’s on YouTube. And it’s the best hour you’ll spend this week if you want to understand why hardware is the true battleground of AI.
Similar Articles
@GoogleDeepMind: Watch → https://goo.gle/4pxlGEh Spotify → https://goo.gle/4f89R2a Apple Podcasts → https://goo.gle/4fpWThL Or listen wh…
Google DeepMind podcast discusses AI interpretability (mechanistic interpretability) and chain-of-thought reasoning, explaining why we need to understand the internal working mechanisms of neural networks and the value and limitations of chain-of-thought as a temporary window.
@vivekgalatage: It's super interesting to know the system architecture of the TPUs. https://henryhmko.github.io/posts/tpu/tpu.html…
A deep dive into Google's TPU architecture, explaining the design philosophy of systolic arrays, pipelining, and ahead-of-time compilation that enables high throughput and energy efficiency.
@unicodef1wn: Ex-Google engineer explained AI agent memory in 12 minutes better than $500 courses. user prompt → working memory → LLM…
An ex-Google engineer explains AI agent memory architecture in 12 minutes, covering working memory and three memory layers (procedural, semantic, episodic) with a summarizer to prevent token bloat, as used by Claude.
@GoogleDeepMind: A model’s chain of thought acts like a scratch pad, offering a window into its reasoning. On the latest episode of our …
Google DeepMind announces a podcast episode on interpretability, featuring host @fryrsquared and @NeelNanda5 discussing mechanistic interpretability, chain of thought monitoring, and safety auditing.
@bookwormengr: Interview with Richard Ho (leader of OpenAI's Jalapeno chip).
In the interview, OpenAI VP of Hardware Richard Ho looks back on his chip career — from Stanford and the D. E. Shaw Anton supercomputer to eight generations of Google TPUs — and reveals the team-building philosophy behind OpenAI's in-house accelerator, Jalapeno: team first, ideas second; a blank-sheet design approach; and, most importantly, the key engineering lesson of keeping hardware programmable and avoiding overfitting to specific models.