Legal LLMs, Pre-Training & Custom AI Hardware

Australian private M&A lawyer turned legal AI founder. Building a startup focused on private equity fund documentation, with custom inference hardware to keep client data off the public cloud. Deep expertise in model pre-training, open-weight LLMs (Qwen, Kimi, DeepSeek), and running AI infrastructure at home. Background: private fund legal work in Sydney and Hong Kong.

Chapters

1. 0:00Intro — AI Tinkerers reunion & meeting Jeremy
2. 1:24Meet Jeremy — M&A lawyer turned hardware geek
3. 2:47From Dogecoin mining to AI inference hardware
4. 3:42Legal AI startup for private equity funds
5. 5:48Running LLM servers from your living room
6. 7:00Space heaters vs graphics cards in a Hong Kong summer
7. 9:05The pre-training data ceiling — scraping the internet isn't enough
8. 11:03Are LLMs the right tool for legal work?
9. 13:43One comma can change everything — legal precision vs AI
10. 16:22Tacit knowledge and the apprenticeship problem
11. 18:55Unitree robots at China's Spring Festival — real-time adaptation
12. 20:32Yeet hay and Chinese medicine — embodied knowledge as big data
13. 23:28Blade Runner's "Tears in Rain" — what data can't capture
14. 26:15Hong Kong media's AI blind spot
15. 27:04A lady using Qwen at Sam's Club in Shenzhen
16. 29:28AI Tinkerers — Hong Kong's practitioner community
17. 30:28Open-weight vs open-source — why Qwen matters
18. 33:54Small models, big punch — the Japanese car analogy
19. 35:55Will AI pricing be sustainable? — the VC funding math
20. 37:25Pier 3 and a love letter to Hong Kong
21. 42:52Trains, borders, and the Greater Bay Area future
22. 48:55Outro — dogs, harbour views & see you next time

About This Episode

Jeremy is a private M&A lawyer who runs a legal AI startup — and runs his own LLM inference hardware out of his living room because he's been building custom PCs since age 14 (for Call of Duty), then mined Dogecoin on graphics cards, and discovered all that knowledge transfers perfectly to training and serving models. We go deep on why lawyers can't use ChatGPT, Claude, or Gemini with client data, why the entire private equity domain is essentially absent from LLM training corpora, why Qwen punches way above its weight, and why the pre-training ceiling is real.