

DeepSeek: Beyond the Headlines | Panel
Feb 25, 2025
In this insightful discussion, guests Veronica Hylák, co-founder of MetaLinguist, AI innovator, Claudio Fantinuoli, CTO at CUDO, and Manuel Herranz, CEO of Pangeanic, delve into groundbreaking advancements in AI and localization. They cover China's emerging startup DeepSeek and its potential to rival OpenAI, highlighting its unique training methods and privacy concerns. The conversation also touches on the geopolitical dynamics of AI investments, ethical implications, and the shift towards self-hosted language models to enhance low-resource language translation.
AI Snips
Chapters
Transcript
Episode notes
DeepSeek's AI Training Innovation
- DeepSeek flipped traditional AI training by prioritizing reinforcement learning over supervised fine-tuning.
- This approach allegedly cut training costs drastically, challenging the billion-dollar norm in AI development.
DeepSeek Sparks Geopolitical AI Race
- DeepSeek triggered a geopolitical AI race among China, the US, and the EU.
- This race will accelerate AI development with significant impact on industries and technology interaction.
Limits of Large Language Models' Reasoning
- LLMs excel at scalable shallow reasoning but lack deep reasoning without structured techniques.
- Humans retain superiority in deep analysis and rich contextual understanding beyond current AI.