Private Equity Funcast Shock and Nawww : DeepSeek and the Future of LLMs
Jan 30, 2025
Join Alan Williamson, CTO of HiBid, and Kristopher Kubicki, CTO of OpenBrand, as they delve into the recent DeepSeek AI announcement and its seismic impact on the tech and VC sectors. The duo debates whether the hype represents a genuine paradigm shift or just noise. They highlight the rapid evolution of AI, particularly LLMs, in shaping software development and investment strategies. The conversation skips between the challenges of AI adoption and the intriguing dynamics between agile teams and larger enterprises in the ever-changing tech landscape.
AI Snips
Chapters
Transcript
Episode notes
DeepSeek's Training Costs
- DeepSeek claimed to train their model using 2048 NVIDIA H800 GPUs for $5.6 million.
- Competitors and Financial Times articles disputed this, suggesting they used more GPUs and potentially distilled data from OpenAI's API.
Testing and Resource Allocation for AI Models
- Test new AI models extensively, especially for specific tasks.
- Allocate resources based on the task's importance: use powerful models for critical tasks and less powerful ones for simpler tasks.
AI Model Selection Challenges
- The abundance of AI models creates a decision-making challenge for users.
- It shifts the focus from building engines to choosing the right engine and managing the data on top.