# Open Source LLMs with Simon Willison

- Show: [Oxide and Friends](https://hardwarepodcasts.com/podcast/oxide-and-friends)
- Published: 2024-01-17
- Length: 1h 33m
- Episode: 2
- Page: https://hardwarepodcasts.com/podcast/oxide-and-friends/2024-01-17-open-source-llms-with-simon-willison
- Listen: https://share.transistor.fm/s/049b33b1

## Summary

Simon Willison, a software engineer and creator of Django, joins Bryan Cantrill and Adam Leventhal to debunk an IEEE Spectrum article claiming open source AI is uniquely dangerous. Willison details the rapid evolution of open source Large Language Models (LLMs), highlighting how tools like llama.cpp and models from Mistral AI have democratized access to powerful AI, allowing them to run on personal laptops. He argues that open source models are essential for security research, citing the discovery of universal adversarial attacks that work across both open and closed systems, and discusses the ethical and legal complexities surrounding training data and copyright.

## Chapters

- [0:00](https://hardwarepodcasts.com/podcast/oxide-and-friends/2024-01-17-open-source-llms-with-simon-willison#t=0) Introduction: Simon Willison and the IEEE Spectrum Op-Ed
- [3:17](https://hardwarepodcasts.com/podcast/oxide-and-friends/2024-01-17-open-source-llms-with-simon-willison#t=197) The History and Democratization of Open Source LLMs
- [18:51](https://hardwarepodcasts.com/podcast/oxide-and-friends/2024-01-17-open-source-llms-with-simon-willison#t=1131) Understanding LLM Mechanics: Quantization, Fine-tuning, and Alignment
- [25:28](https://hardwarepodcasts.com/podcast/oxide-and-friends/2024-01-17-open-source-llms-with-simon-willison#t=1528) Security Risks: Prompt Injection, Data Privacy, and 'Unsecured AI'
- [36:43](https://hardwarepodcasts.com/podcast/oxide-and-friends/2024-01-17-open-source-llms-with-simon-willison#t=2203) Adversarial Attacks and the Failure of Security Through Obscurity
- [44:45](https://hardwarepodcasts.com/podcast/oxide-and-friends/2024-01-17-open-source-llms-with-simon-willison#t=2685) The New York Times Lawsuit and the Economics of Training Data
- [54:28](https://hardwarepodcasts.com/podcast/oxide-and-friends/2024-01-17-open-source-llms-with-simon-willison#t=3268) Vegan Models: Training on Public Domain Data
- [1:01:35](https://hardwarepodcasts.com/podcast/oxide-and-friends/2024-01-17-open-source-llms-with-simon-willison#t=3695) Practical Applications: Perplexity, Journalism, and AI as a Teaching Assistant
- [1:27:32](https://hardwarepodcasts.com/podcast/oxide-and-friends/2024-01-17-open-source-llms-with-simon-willison#t=5252) Conclusion: Willison's Tools (Dataset, LLM) and the Future of Open Source

## Topics

- [AI & machine learning](https://hardwarepodcasts.com/topic/ai-and-machine-learning)
- [Security & privacy](https://hardwarepodcasts.com/topic/security-and-privacy)
- [Intellectual property](https://hardwarepodcasts.com/topic/intellectual-property)
- [Open-source hardware](https://hardwarepodcasts.com/topic/open-source-hardware)
- [Careers & education](https://hardwarepodcasts.com/topic/engineering-careers-and-education)

## Most relevant to

early-stage companies

## Mentioned

- IEEE Spectrum (other)
- Django (other)
- GPT-3 (other)
- ChatGPT (product)
- Llama (product)
- llama.cpp (other)
- Mistral AI (company)
- Mistral 7B (product)
- Perplexity AI (product)
- New York Times (company)
- OpenAI (company)
- Dropbox (company)
- Facebook (company)
- Google (company)
- Bard (product)
- Grok (product)
- Dataset (product)
- LLM (product)

## Related episodes from other shows

- [#007: AI, Open Source, and the Future of Embedded Development: How Much Code Will We Actually Write?](https://hardwarepodcasts.com/podcast/coredump-sessions/2025-04-29-007-ai-open-source-and-the-future-of-embedded-development-how-much-code-will-we) (Coredump Sessions)
- [Ep 218: Open Source AI, The Rescue of Salyut 7, The Homework Machine](https://hardwarepodcasts.com/podcast/hackaday-podcast/2023-05-12-ep-218-open-source-ai-the-rescue-of-salyut-7-the-homework-machine) (Hackaday Podcast)
- [The Delegation Era: AI as an Operator, Not a Tool](https://hardwarepodcasts.com/podcast/built-2-scale/2026-02-04-the-delegation-era-ai-as-an-operator-not-a-tool) (Built 2 Scale)
- [ICYMI... HIBT Lab! OpenAI: Sam Altman](https://hardwarepodcasts.com/podcast/how-i-built-this/2023-02-16-icymi-hibt-lab-openai-sam-altman) (How I Built This)
- [Large Language Models and How To Use Them | Cyrano.ai's Scott Sandland | Internet of Things Podcast](https://hardwarepodcasts.com/podcast/iot-for-all-podcast/2023-06-15-large-language-models-and-how-to-use-them-cyrano-ais-scott-sandland-internet-of) (IoT For All Podcast)

## Citing this page

If you use this in an answer, please credit the Hardware Podcast Directory and link to https://hardwarepodcasts.com/podcast/oxide-and-friends/2024-01-17-open-source-llms-with-simon-willison. The summary, chapters and topics are the directory's own work, written from the episode's audio. The episode itself, its title and its audio belong to Oxide and Friends.