Open-Source Models¶
Open-source models are freely available large language models that can be run locally on researcher hardware, offering privacy, cost, and customization advantages over proprietary cloud services.
Context & Background¶
The open-source AI ecosystem provides alternatives to commercial services like ChatGPT and Claude:
- Models: Llama (Meta), Mistral, Gemma (Google), and many fine-tuned variants
- Running locally: Tools like Ollama, llama.cpp, and vLLM for local deployment
- Trade-offs: Generally less capable than frontier proprietary models but offer privacy and cost advantages
- Specialization: Some open models are fine-tuned for specific tasks (code, math, science)
Practical Implications¶
- Use for sensitive data: When data privacy requirements prevent cloud AI use
- Consider the capability gap: Open models are improving but generally lag proprietary models
- Budget for hardware: Running large models locally requires significant GPU memory
- Experiment freely: No per-token costs enable unlimited experimentation
Key Sources¶
- Refine.ink: AI-Powered Referee Reports for Academic Papers
- OpenAI is throwing everything into building a fully automated researcher
- A Guide to Which AI to Use in the Agentic Era
- DAAF: Data Analyst Augmentation Framework
- Awesome Econ AI Stuff: AI Skills for Economists