
TBPN Meet the New Thiel Fellows, GPT 5.5, Thoma Bravo Loses Medallia | Victor Boyd, Alex Shieh, Nick Dobroshinsky, Ishan Gupta, Antoni Kiszka, Milan Lustig, Galen Mead, Aubrey Niederhoffer, Samuel Carvalho, Claire Wang
93 snips
Apr 23, 2026 Victor Boyd, Cavalla CTO building autonomous forklifts, talks full-stack warehouse robots, failed retrofits, and Xbox-controller teleoperation. Alex Shieh, The Antifraud Company co-founder and ex-Brown student, covers AI fraud hunting, OSINT, and legal recoveries. Milan Lustig, Opt32 co-founder focused on robotics compute, gets into software-to-chip infrastructure. Plus GPT-5.5, Medallia drama, and LLM stock research.
AI Snips
Chapters
Books
Transcript
Episode notes
Robotics Needs A Compute Stack Built For Constraints
- Milan Lustig argues physical autonomy needs full-stack compute infrastructure because robots face brutal latency, power, and cost limits that cloud-scale chips ignore.
- Opt32 starts with model optimization software on existing NVIDIA hardware, then plans to descend toward custom boards, FPGAs, and eventually ASICs.
AI Pushes Computer Science Up The Abstraction Ladder
- Milan Lustig thinks AI will not kill computer science so much as push valuable CS work up the abstraction ladder.
- He expects system architecture and theory to matter more while low-level implementation jobs like basic front-end work lose importance.
Computer Use Models Turn Screens Into Training Data
- Galen Mead views computer use as a general action model because screens and keyboards are the universal interface for real-world digital work.
- Standard Intelligence trains on video inputs and outputs mouse deltas plus character-level keystrokes, treating screen recordings as agent pretraining data.






