Getting Started With Open Models
A practical guide to commodity intelligence, owning your AI stack, and deciding when to still reach for the frontier
TL;DR
- Frontier AI models have advanced but become more expensive and restricted.
- Cheaper, open-weight AI models are now competitive for routine tasks, offering cost savings and independence.
- Choosing the right model depends on the task's complexity, vagueness, and the consequences of errors.
- Open-weight models make model weights available for download, allowing users to run them independently.
- Most powerful open models currently originate from Chinese labs, used for marketing and competitive pressure.
- Owning your AI stack involves choosing models, hosting, and integration layers, with options for self-hosting or using providers.
- Tasks suitable for open models are often routine, easily verifiable, and have clear inputs and outputs.
- A systematic audit process helps identify tasks that can be transitioned from frontier AI to commodity intelligence.
- The guide recommends starting with small, reversible experiments to test suitability before full automation.
- When evaluating candidates, consider recurrence, input clarity, output definition, verifiability, stakes, economics, and integration with existing habits.