What are the key takeaways from this Hard Fork episode?

Key insights include: **Performance leap:** Gemini 3 Pro scores 37.5% on Humanity's Last Exam benchmark versus 21.6% for Gemini 2.5 Pro, showing substantial improvement on graduate-level interdisciplinary questions across multiple evaluation metrics.; **Dynamic interface generation:** The model creates custom interactive interfaces for queries, building tutorials with images and interactive elements or coding mortgage calculators, moving beyond standard text responses to personalized user experiences.; **AGI timeline unchanged:** Hassabis maintains his five to ten year AGI prediction, stating Gemini 3 progress aligns with expectations but requires one or two additional research breakthroughs in reasoning, memory, and world modeling capabilities.

What did Demis Hassabis and Josh Woodward discuss on Hard Fork?

Google releases Gemini 3 Pro, demonstrating significant benchmark improvements over previous models. DeepMind CEO Demis Hassabis and VP Josh Woodward discuss new capabilities, competitive positioning, and timeline toward artificial general intelligence. Key topics include: **Performance leap:** Gemini 3 Pro scores 37.5% on Humanity's Last Exam benchmark versus 21.6% for Gemini 2.5 Pro, showing substantial improvement on graduate-level interdisciplinary questions across multiple evaluation metrics.; **Dynamic interface generation:** The model creates custom interactive interfaces for queries, building tutorials with images and interactive elements or coding mortgage calculators, moving beyond standard text responses to personalized user experiences..

How long is this episode of Hard Fork?

This episode is 26 minutes long. SignalCast provides an AI-generated summary so you can get the key insights in about 3 minutes.

Hard Fork

Google's Gemini 3 Is Here: A Special Early Look

November 18, 2025

26 min episode · 2 min read

Demis Hassabis,Josh Woodward

Episode

26 min

Read time

2 min

Topics

Investing, Fundraising & VC, Leadership

AI-Generated Summary

Published Dec 22, 2025

Key Takeaways

✓Performance leap: Gemini 3 Pro scores 37.5% on Humanity's Last Exam benchmark versus 21.6% for Gemini 2.5 Pro, showing substantial improvement on graduate-level interdisciplinary questions across multiple evaluation metrics.
✓Dynamic interface generation: The model creates custom interactive interfaces for queries, building tutorials with images and interactive elements or coding mortgage calculators, moving beyond standard text responses to personalized user experiences.
✓AGI timeline unchanged: Hassabis maintains his five to ten year AGI prediction, stating Gemini 3 progress aligns with expectations but requires one or two additional research breakthroughs in reasoning, memory, and world modeling capabilities.
✓Distribution advantage: Google integrates Gemini 3 into products serving billions of daily users including search AI mode, leveraging existing monopoly position to generate usage data and model improvements competitors cannot match.

What It Covers

Google releases Gemini 3 Pro, demonstrating significant benchmark improvements over previous models. DeepMind CEO Demis Hassabis and VP Josh Woodward discuss new capabilities, competitive positioning, and timeline toward artificial general intelligence.

Key Questions Answered

•Performance leap: Gemini 3 Pro scores 37.5% on Humanity's Last Exam benchmark versus 21.6% for Gemini 2.5 Pro, showing substantial improvement on graduate-level interdisciplinary questions across multiple evaluation metrics.
•Dynamic interface generation: The model creates custom interactive interfaces for queries, building tutorials with images and interactive elements or coding mortgage calculators, moving beyond standard text responses to personalized user experiences.
•AGI timeline unchanged: Hassabis maintains his five to ten year AGI prediction, stating Gemini 3 progress aligns with expectations but requires one or two additional research breakthroughs in reasoning, memory, and world modeling capabilities.
•Distribution advantage: Google integrates Gemini 3 into products serving billions of daily users including search AI mode, leveraging existing monopoly position to generate usage data and model improvements competitors cannot match.

Notable Moment

Hassabis acknowledges parts of the AI industry show bubble characteristics, citing seed funding rounds reaching tens of millions with minimal substance, while defending Google's diversified approach across immediate products and long-term moonshots.

Know someone who'd find this useful?