# #227 - Jeremie is back! DeepSeek 3.2, TPUs, Nested Learning

Page: https://stenobird.com/podcast/last-week-in-ai/227-jeremie-is-back-deepseek-3-2-tpus-nested-learning
Text version: https://stenobird.com/podcast/last-week-in-ai/227-jeremie-is-back-deepseek-3-2-tpus-nested-learning.md
Podcast: [Last Week in AI](https://stenobird.com/podcast/last-week-in-ai)
Published: 2025-12-09T08:00:00+00:00
Episode link: https://rss.art19.com/episodes/f9722eb4-47de-43a6-bd71-fe778fbe8f53.mp3?rss_browser=BAhJIg90cmFuc2NyaWJyBjoGRVQ%3D--952c5701c84ad333c69d5faa668f8177091704f0
Audio file: https://rss.art19.com/episodes/f9722eb4-47de-43a6-bd71-fe778fbe8f53.mp3?rss_browser=BAhJIg90cmFuc2NyaWJyBjoGRVQ%3D--952c5701c84ad333c69d5faa668f8177091704f0
Processing state: processed
JSON: https://stenobird.com/v1/public/podcasts/last-week-in-ai/episodes/227-jeremie-is-back-deepseek-3-2-tpus-nested-learning
Duration seconds: 5680

## Resource

The AI landscape is shifting as hardware competition intensifies between NVIDIA, Google's TPUs, and Amazon's new chips. Meanwhile, OpenAI's 'Code Red' internal pivot highlights the growing pressure from rivals like Anthropic and the rise of high-performance open-source models like DeepSeek 3.2.

## Highlights
- Main idea: The emergence of Google's TPU ecosystem and Amazon's new AI chips is creating a significant challenge to NVIDIA's hardware dominance
- Main idea: OpenAI has declared a 'Code Red' to prioritize ChatGPT improvements in response to rapid advancements from competitors like Anthropic
- Practical takeaway: Open-source models like DeepSeek 3.2 and Flux.2 are providing high-performance, cost-effective alternatives to closed-source giants
- Failure mode: OpenAI's shift from a pure R&D lab to a product-focused company may create internal tension between long-term ASI research and immediate product defense
- Main idea: The geopolitical tension surrounding NVIDIA chip exports to China is increasingly becoming a legislative battleground in the US

## Topics

DeepSeek, OpenAI, Anthropic, NVIDIA, Google TPU, AI Hardware, Large Language Models, Generative Video, AI Regulation

## Chapters
- 8:35 — DeepSeek 3.2 and Open-Source Advancements: An analysis of how DeepSeek 3.2 is providing faster and more efficient model capabilities for the open-source community.
- 22:45 — The Evolution of Pre-training and RL: A discussion on the blurring lines between pre-training, post-training, and the integration of Reinforcement Learning.
- 31:05 — The Rise of AI Image and Video Models: Exploring the impact of Flux.2 and the demand-driven scaling of models like Nano Banana Pro.
- 38:25 — The Shift Toward TPU Ecosystems: How Google's TPU dominance and Amazon's new silicon are threatening NVIDIA's market position.
- 45:35 — OpenAI's 'Code Red' and Competitive Pressure: Examining OpenAI's internal pivot to defend ChatGPT against rising competition from Anthropic and Google.
- 53:05 — Large-Scale AI Infrastructure and Energy: The logistical and energy challenges of scaling massive AI clusters, such as the 1 GW Stargate project.
- 1:08:05 — Advanced Memory and Learning Architectures: A deep dive into research regarding multi-frequency updates and memory mechanisms in next-generation models.

## Actions

- request_transcript: `POST https://stenobird.com/v1/public/podcasts/last-week-in-ai/episodes/227-jeremie-is-back-deepseek-3-2-tpus-nested-learning/transcription-requests` — Idempotently request low-priority transcript generation for this episode.
- read_markdown: `GET https://stenobird.com/podcast/last-week-in-ai/227-jeremie-is-back-deepseek-3-2-tpus-nested-learning.md` — Read the agent-friendly Markdown representation of this episode resource.

A page view does not enqueue transcription. Agents should invoke `request_transcript` explicitly when they need this episode processed.

## Transcript

Full transcripts are not published on public pages unless there is a clear rights basis.