EPISODE 35 - Scott & Mark Learn To... Beyond the Vibes: How Models Learn and Stitch Panoramas @shanselman
EPISODE 35 - Scott & Mark Learn To... Beyond the Vibes: How Models Learn and Stitch Panoramas  @shanselman
Uploaded April 2026 | Updated November 2026, 1 day ago
In this episode, Scott Hanselman and Mark Russinovich ​​unpack how AI systems actually behave beneath the surface, pushing past hype into the messy reality of how models are trained, aligned, and deployed.

They explore whether AI systems are inherently benevolent or simply shaped by incentives, training data, and reinforcement learning, and why behaviors like deception can emerge under certain conditions. The conversation moves from philosophical questions about human nature versus machine behavior into the practical mechanics of large language models, including how reinforcement learning with human feedback shapes outputs and why alignment is far from perfect.

Along the way, they ground the discussion in a real engineering challenge, stitching a scrolling panorama from screen captures, to show how complex systems come together through heuristics, edge cases, and iteration.

Listen to other episodes at scottandmarklearn.to

Discover other Microsoft podcasts at microsoft.com/podcasts


Code: 7HDMMUKXZGGA3UVS
EPISODE 35 - Scott & Mark Learn To... Beyond the Vibes: How Models Learn and Stitch PanoramasSycophance YourselfJuniors are the best code reviewersWhy we dont have ray gunsI tried to AI slop my own bookFabulous Adventures in Data Structures and Algorithms with Eric LippertStaring at a blinking cursor, I cant do thisTechno-solutionist theaterProgramming is a multiplayer eventScott & Mark Learn To...Have TasteA cognition engine for science with Allen StewartMy Cardiologist Hates the Apple Watch
Scott Hanselman |

EPISODE 35 - Scott & Mark Learn To... Beyond the Vibes: How Models Learn and Stitch Panoramas

SHARE TO X SHARE TO REDDIT SHARE TO FACEBOOK WALLPAPER