In this episode of ODSC’s AiX Podcast, we sit down with Sinan Ozdemir, mathematician, AI expert, and founder ofLoopGenius, to break down the hype vs. reality surrounding DeepSeek R1. In just a few short weeks, this open-source reasoning model has upended expectations—rivaling OpenAI’s O1, triggering stock market reactions, and sparking a wave of rapid AI releases from industry giants.

Together, we explore whether DeepSeek R1 is truly a paradigm shift in AI or simply the next iteration in an increasingly fast-moving race. We also tackle the controversy over model distillation, the impact on AI startups, and what this means for the future of open-source development and reasoning models.

Key Takeaways 

🔹“The big story isn’t just that DeepSeek R1 is as good as OpenAI’s model—it’s who made it.”

🔹"We shouldn’t just assume reasoning is better. Sometimes, a model talks itself out of the right answer”

🔹“Mixture of Experts? Reinforcement Learning? None of these techniques are new. It’s just a wake-up call that OpenAI isn’t the only one doing this.”

🔹"Inference costs are the real bottleneck. Open-source doesn't mean free—it means you pay for hosting it."

 🔹"The price of training AI is falling, but the Jevons paradox applies—the cheaper AI gets, the more we’ll use it."

🔹"We might need a blind taste test for LLMs—because once you strip away branding, most users wouldn’t know the difference."

Topics DiscussedDeepSeek R1 & AI’s Competitive Landscape

Why DeepSeek’srelease triggered a market shake-up

OpenAI’s rapid response: O3-mini and the rise ofAI agentic workflows

Google’s Gemini 2.0 and thearms race in AI reasoning models

Mixture of Experts (MoE): How DeepSeek is optimizing compute efficiency

Reinforcement Learning with No Human Feedback (RL-NHF): The self-improving approach to training

Distillation Controversy: AI models trained on outputs from other models

How DeepSeek R1 isreshaping the open-source vs. proprietary debate

Can DeepSeek’s cost-efficient approachunlock new AI opportunities for startups?

Thegeopolitical implications of China’s rapid AI advancements

Technical Deep Dive: What Makes DeepSeek R1 Unique?The Future of Open-Source AI

References & Resources Mentioned

DeepSeek R1 Technical Report: https://github.com/deepseek-ai/DeepSeek-R1/blob/main/DeepSeek_R1.pdf

The Mixture of Experts Model (MoE) Explained:⁠⁠https://huggingface.co/blog/moe

LLM Distillation:⁠https://snorkel.ai/blog/llm-distillation-demystified-a-complete-guide⁠

Chain-of-Thought:⁠https://en.wikipedia.org/wiki/Prompt_engineering⁠

DeepSeek’s API & Open Model Access:⁠ DeepSeek AI⁠

Reinforcement Learning:https://en.wikipedia.org/wiki/Reinforcement_learning

The Jevons Paradox in AI Compute Demand: ⁠https://www.npr.org/sections/planet-money/2025/02/04/g-s1-46018/ai-deepseek-economics-jevons-paradox⁠

ODSC East 2025 coming up May 13th-15th in Boston:https://odsc.com/boston/

This episode was sponsored by:

🎤ODSC East 2025 – The Leading AI Builders Conference –⁠ https://odsc.com/boston/⁠Join us from May 13th to 15th in Boston forhands-on workshops, training sessions, and cutting-edge AI talks coveringgenerative AI, LLMOps, and AI-driven automation.

🔔Never miss an episode—subscribe now!