DeepSeekMath-V2: Self-Verifiable Math Reasoning

Our 227th episode with a summary and discussion of last week's big AI news!

Recorded on 12/05/2025

Hosted by Andrey Kurenkov and Jeremie Harris

Feel free to email us your questions and feedback at contact@lastweekinai.com and/or hello@gladstone.ai

Read out our text newsletter and comment on the podcast at https://lastweekin.ai/

In this episode:

• Deep Seek 3.2 and Flux 2 release, showcasing advancements in open-source AI models for natural language processing and image generation respectively.
• Amazon's new AI chips and Google's TPUs signal potential shifts in AI hardware dominance, with growing competition against Nvidia.
• Anthropic's potential IPO and OpenAI's declared ‘Code Red’ indicate significant moves in the AI business landscape, including high venture funding rounds for startups.
• Key research papers from DeepMind and Google explore advanced memory architectures and multi-agent systems, indicating ongoing efforts to enhance AI reasoning and efficiency.

Timestamps:

• (00:00:10) Intro / Banter
• (00:02:42) News Preview
• Tools & Apps
• (00:03:30) Deepseek 3.2 : New AI Model is Faster, Cheaper and Smarter
• (00:23:22) Black Forest Labs launches Flux.2 AI image models to challenge Nano Banana Pro and Midjourney
• (00:28:00) Sora and Nano Banana Pro throttled amid soaring demand | The Verge
• (00:29:34) Mistral closes in on Big AI rivals with new open-weight frontier and small models | TechCrunch
• (00:31:41) Kling's Video O1 launches as the first all-in-one video model for generation and editing
• (00:34:07) Runway rolls out Gen 4.5 AI video model that beats Google, OpenAI
• Applications & Business
• (00:35:18) NVIDIA’s Partners Are Beginning to Tilt Toward Google’s TPU Ecosystem, with Foxconn Reportedly Securing TPU Rack Orders
• (00:40:37) Amazon releases an impressive new AI chip and teases an Nvidia-friendly roadmap | TechCrunch
• (00:43:03) OpenAI declares ‘code red’ as Google catches up in AI race | The Verge
• (00:46:20) Anthropic reportedly preparing for massive IPO in race with OpenAI: FT
• (00:48:41) Black Forest Labs raises $300M at $3.25B valuation | TechCrunch
• (00:49:20) Paris-based AI voice startup Gradium nabs $70M seed | TechCrunch
• (00:50:10) OpenAI announced a 1 GW Stargate cluster in Abu Dhabi
• (00:53:22) OpenAI’s investment into Thrive Holdings is its latest circular deal
• (00:55:11) OpenAI to acquire Neptune, an AI model training assistance startup
• (00:56:11) Anthropic acquires developer tool startup Bun to scale AI coding
• (00:56:55) Microsoft drops AI sales targets in half after salespeople miss their quotas - Ars Technica
• Projects & Open Source
• (00:57:51) [2511.22570] DeepSeekMath-V2: Towards Self-Verifiable Mathematical Reasoning
• (01:01:52) Evo-Memory: Benchmarking LLM Agent Test-time Learning with Self-Evolving Memory
• Research & Advancements
• (01:05:44) Nested Learning: The Illusion of Deep Learning Architecture
• (01:13:30) Multi-Agent Deep Research: Training Multi-Agent Systems with M-GRPO
• (01:15:50) State of AI: An Empirical 100 Trillion Token Study with OpenRouter
• Policy & Safety
• (01:21:52) Trump signs executive order launching Genesis Mission AI project
• (01:24:42) OpenAI has trained its LLM to confess to bad behavior | MIT Technology Review
• (01:29:34) US senators seek to block Nvidia sales of advanced chips to China

See Privacy Policy at https://art19.com/privacy and California Privacy Notice at https://art19.com/privacy#do-not-sell-my-info.