Hi Dotikers!

Before diving into the news, a quick note: we're taking a week off. No newsletter in your inbox next week. We're thinking of all our subscribers, but even AI enthusiasts need to recharge their batteries. See you in January, fresh and ready to go 🙂

After Nvidia's 20 billion for Groq yesterday, we're heading backstage with a leak that's pure gold for prompt engineering enthusiasts. Jane Manchun Wong, the reverse engineer known for digging through app internals, unearthed the complete system prompt for the Gemini assistant that Waymo is preparing for its robotaxis. And this isn't some footnote: we're talking 1,200 lines of rules, limits, and codified behaviors.

What stands out first is the sharp separation between the conversational assistant and the autonomous driving system. Gemini can change the temperature, skip to the next song, turn on cabin lights. But touch the steering wheel, even virtually? Forbidden. If a passenger asks to speed up or change the route, the AI must politely deflect. The prompt goes even further: if someone mentions a crash video involving Waymo, Gemini must neither confirm, deny, nor speculate. You can feel the legal team's fingerprints on every line.

The document also reveals pre-approved jokes, dad jokes calibrated to offend no one. Waymo even thought about the psychology of anxious passengers: reassuring phrases are ready to calm those who freak out about riding without a driver. This is trust engineering at its finest.

While Tesla is betting on Grok for a chattier, more laid-back companion, Waymo is playing the sober and functional assistant card. This leak is a goldmine for anyone interested in AI product design. Between the lines, you can read all the anxieties of a manufacturer facing a technology the general public hasn't fully embraced yet.

G.