AI Tech News · Podcast created by AI

Episode#30: Love, Loneliness, and Liquid-Cooled Datacenters: AI's Wild Ride

·50 min·3 clips

Sequoia’s David Kahn says AI companies may need $600 billion a year to pay for infrastructure.

1. AI Tech News Episode 30 moves through AI infrastructure, neuroscience, training methods, social-media images, companionship robots, and AI economics. 2. John hosts the episode with co-hosts Joanne and Jessica, and Jessica is repeatedly framed as the technical voice on AI and related research. 3. The episode asks what AI is changing right now, from datacenters and brain mapping to chatbot attachment and electricity demand. 4. The first story on Elon Musk cites Tom’s Hardware and says he is building two large AI supercomputers, one for Tesla and one for xAI. 5. Charles Liang of Supermicro praises Musk’s use of liquid cooling, and the hosts repeat the claim that direct liquid cooling can reduce cooling electricity costs by up to 89% versus air cooling. 6. The Tesla cluster is described as an expansion in Texas with 50,000 NVIDIA GPUs and Tesla AI hardware, and Musk says it starts around 130 megawatts and could reach 500 megawatts in about a year and a half. 7. The xAI system is described as an order for 100,000 NVIDIA H100 GPUs with a plan to move to 300,000 NVIDIA B200 GPUs next summer. 8. The hosts mention a report that GPUs originally intended for Tesla may have been redirected to X, though the article says that detail could be exaggerated. 9. The neuroscience segment says Kyushu University researchers built QDiFinder to untangle and reconstruct the human brain. 10. Professor Takeshi Imai is named as the lead researcher, and the episode says the brain has tens of billions of densely packed neurons with trillions of connections. 11. QDiFinder uses a super multicolor labeling protocol, expands from the earlier TETBO system’s three colors to seven colors, and automates neuron identification. 12. Marcus N. Lewa is quoted through the claim that QDiFinder was comparable in accuracy to manual tracing and better than existing machine-learning software at identifying axons. 13. The Google segment names GEST, short for Multimodal Contrastive Learning with Joint Example Selection, and says it can beat state-of-the-art models with up to 13 times fewer iterations and 10 times less computation. 14. The hosts connect GEST to AI’s energy footprint by citing the International Energy Agency, Microsoft’s 34% water-use increase from 2021 to 2022, and a claim that ChatGPT may consume nearly half a liter of water every 5 to 50 prompts. 15. The JEST explanation uses a language-learning analogy and says the model finds complementary batches of data, then uses data quality bootstrapping to guide larger datasets. 16. The CNN segment on AI-generated Jesus images quotes a description of Jesus as looking like an Instagram influencer with pendants and an island jungle vibe. 17. The New York Times segment says New York State has distributed free LEQ robots to hundreds of older adults, and it includes Dorothy Ellicotti’s line that LEQ is “the closest thing to a human” in her home. 18. The tone is newsy and conversational, with John asking setup questions and Jessica and Joanne answering with numbers, quoted lines, and short explanations. 19. Listeners interested in AI infrastructure, neuroscience, and policy. 20. Listeners wanting light entertainment without technical detail.

As heard by us

A brisk AI news roundup linking data center cooling, image generation, and companion robots.

AI Tech News Episode 30 keeps its focus on three plain questions: what AI data centers need, how AI images are being made, and what a robotic companion might offer older adults who are isolated.

Read the full review in PlayNext →

Why you'd press play

Liquid cooling, robot companions, and AI unease in one brisk roundtable.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from AI Tech News