Skip to main content
  1. Blog/

What Two AI Models Concluded From My Accident's Dashcam Footage

·1 min
Author
Chengyu
I’m Chengyu — a final-year Computer Science student at the University of Sydney. I write about the things I build and break, plus hiking, travel, gaming, and gadgets.

After getting the car back, I sent the dashcam footage to Tesla’s engineering team, and separately, out of curiosity, asked two AI models — Gemini Pro and ChatGPT 5.2 — to do their own engineering-style analysis of whether the emergency braking system had been slow to react.

Both models reached a similar conclusion. This was a classic “ghost probe” scenario — a vehicle suddenly appearing from a blind spot — where the other car became visible less than 5 meters away, leaving less than 0.5 seconds of physical reaction time. At roughly 35–39 km/h, the required stopping distance (around 11–16 meters) was far longer than the space actually available (under 10 meters). Neither model found clear evidence that the emergency braking system had been delayed — instead, both concluded the situation was close to a physical limit where a collision was nearly unavoidable no matter how the system responded.

My takeaway: autonomous driving isn’t a simple pass/fail system — it grows by continuously bumping up against the edges of what’s physically possible. Drive carefully, slow down when you can, and put safety first.

Related