In this episode, we welcome Ivan Lee, founder and CEO of Datasaur.ai, a leading expert in LLM and AI solutions, particularly focusing on large language and reasoning models. Today, we delve deep into reasoning models, their recent advancements, real-world use cases, and the practical considerations enterprises should be aware of when adopting these powerful tools.

Whether you're a data scientist, ML engineer, practitioner, or business leader, this conversation offers valuable insights into one of AI’s most dynamic and impactful areas.

Key Topics Covered

• Understanding Reasoning Models – How they differ from traditional large language models by explicitly demonstrating a chain of thought. 
• Recent Developments – Exploring breakthroughs like OpenAI’s O1 and O3-mini, DeepSeek R1, Anthropic’s Claude 3.7 Sonnet, Google’s Gemini Flash Thinking, and Alibaba’s Qwen-32B. 
• Real-World Applications – Where reasoning models excel, such as strategic research, complex analytics, and nuanced decision-making scenarios.
• Challenges and Limitations – Addressing cost, latency, complexity, and non-determinism in practical deployments. 
• Fine-Tuning Reasoning Models – Current challenges, techniques, and the potential of closed-beta programs. 
• Agentic Workflows – Using reasoning models within agentic systems, their strengths, weaknesses, and realistic deployment considerations.
• Future of Reasoning Models – The shift towards specialized, domain-specific models and ongoing commoditization of technology.

Memorable Quotes

💬 "These reasoning models unlock a whole new set of use cases that we didn't believe traditional foundation models would be able to address previously. But the downside is cost and latency”

💬 "A couple of weeks ago, we saw the releases of both Cloud 3.7 as well as OpenAI's GPT 4.5. And what was important about those announcements was as much about what they offered as what they didn't offer. 

💬 "I think a lot of people have been very hyped for a long time to see OpenAI's 4.5 model. Now that we got our hands on it, it really only demonstrated an incremental improvement on what we've seen previously.”

💬 "What could we accomplish if the model is really thinking hard about a problem for longer durations”

Resources & Tools Mentioned

• Datasaur.ai Platform:⁠ www.datasaur.ai⁠
• Ivan Lee on LinkedIn:⁠ linkedin.com/in/ivanlee⁠
• Anthropic’s Claude 3.7 Sonnet: https://www.anthropic.com/news/claude-3-7-sonnet
• OpenAI’s O1 and O3-mini:⁠ ⁠https://openai.com/index/openai-o3-mini/
• Open AI’s Operator: ⁠https://openai.com/index/introducing-operator/⁠
• DeepSeek R1:⁠ deepseek.com⁠
• Google's Gemini Flash Thinking: https://deepmind.google/technologies/gemini/flash-thinking/
• Alibaba's Qwen-32B Model: https://qwenlm.github.io/blog/qwq-32b/
• "Thinking, Fast and Slow" by Daniel Kahneman: ⁠https://en.wikipedia.org/wiki/Thinking,_Fast_and_Slow⁠
• Manus AI - Universal AI Agent platform mentioned in the context of agentic workflows: ⁠https://www.manus.ai⁠
• OpenAI Operator (Agentic system for web-based tasks: ⁠https://openai.com/blog/operator⁠
• Deep Research by OpenAI:⁠ https://openai.com/blog/deep-research⁠Tencent's Huanyuan Turbo S - a new model allowing fast and slow thinking: https://ai.tencent.com
• Firecrawl AI - Headless browser tool for data collection: ⁠ https://github.com/mendableai/firecrawl⁠
• Browserbase - another headless browser tool explicitly: ⁠https://browserbase.com⁠DeepSeek Open Source Week: https://deepseek.com/blog

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 for hands-on workshops, training sessions, and cutting-edge AI talks covering generative AI, LLMOps, and AI-driven automation.

🔥 Use the exclusive code ODSCPodcast for an additional 10% off any ticket type!