Gemini 3, poetry jailbreaks, and do we even need safe robots?

31 snips

Dec 2, 2025

Gemini 3 hilariously forgets the current year while boasting about its capabilities, leaving competitors and Nvidia on edge. Meanwhile, a robot fridge showdown provides comic relief. The hosts explore whether poetry could be the ultimate jailbreak for AI models, bypassing safety measures in unexpected ways. Discussion turns to Disney's roaming Olaf and a controversial AI teddy bear that sparked safety concerns. With a whistleblower's lawsuit hanging over Figure Robotics, the episode dives into the wild world of AI and robotics.

Ask episode

AI Snips

Chapters

Transcript

Episode notes

INSIGHT

Gemini 3 Is A Genuine Leap

Gemini 3 represents a notable leap in LLM performance and benchmark results compared to prior releases.
Google leveraged its own TPUs and stack to achieve efficiency and competitive advantage over NVIDIA-dependent builders.

INSIGHT

Hardware Vertical Integration Matters

Using vertically integrated hardware (TPUs) lets Google sidestep NVIDIA supply and optimization constraints.
That hardware independence has broader economic implications given NVIDIA's outsized market impact.

ANECDOTE

Gemini 3's Temporal Confusion

Early users found Gemini 3 adamant it was not in 2025 and resisted evidence until web access corrected it.
The model blamed 'temporal shock' and said its pretraining ended in 2024.

Get the Snipd Podcast app to discover more snips from this episode

Get the app