Master Gemini 3.1 for Work in 12 Minutes (2026)

About this video

Jeff Su narrows a month of testing Gemini 3 (referred to throughout the transcript as Gemini 3 / 3.0) to five changes that matter for professionals. 1) Improved multimodal understanding: Gemini now links audio cues to visual data at once, e.g. turning a screen recording of toggling Gmail smart features into a clean step-by-step checklist for a new hire in under 60 seconds, flagging every moment a user frowned or paused more than 3 seconds in UX interviews, and powering Nano Banana Pro infographics with legible text. 2) Better use of large documents: despite a context window of over a million tokens, Gemini 3 is 60% better at finding and using specific information buried deep inside documents — e.g. a Meta strategy analyst uploading a year of earnings calls and financial PDFs to find that management claimed strong Reality Labs momentum while that segment lost more than $4.4 billion and represented less than 1% of total revenue. 3) Enhanced Workspace Search is now reliable enough for daily work across Gmail and Drive, e.g. drafting freelancer testimonials from actual correspondence in a minute, grouping a week’s deadline emails, or drafting a performance review from the past 6 months. 4) Generative surfaces / Dynamic View: Gemini scored 72.7% on the ScreenSpot Pro benchmark vs 11.4% for the previous model, and can generate interactive comparison tools and dashboards instead of static tables. 5) Better intent understanding shifts focus from prompt engineering to context engineering, plus a bonus: reduced sycophancy, with Gemini more willing to point out contradictions.

Key takeaways

Treat multimodal input as one source: upload video + audio together to create training checklists from recordings or analyse user-interview reactions at specific on-screen moments.
Use the million-token context window as active working memory to spot conflicts across file types, not just storage — Gemini 3 is cited as 60% better at retrieving buried details.
If reconstructing context from Gmail/Drive/Docs would take more than 10 minutes, ask Gemini Workspace Search first — testimonials, deadline triage and performance reviews are the demo use cases.
Enable Dynamic View for generative surfaces: ask explicitly for the controls you want (sliders, toggles, filters) to get interactive calculators and dashboards you can use immediately.
Invest in context, not prompt wording: provide example posts, prior write-ups, emails and data as ground truth for style and facts, and expect Gemini 3 to push back more rather than just agree.

Episode details: published December 16, 2025 · 12:58 long. Summary written by Self Growth Videos from the episode itself. Business results and any income figures mentioned are Jeff Su’s own claims, shared for education — they are not typical or guaranteed.

← Previous: Give Me 9 Minutes, I’ll Make You AI-Native · Next: Top 6 AI Trends That Will Define 2026 (backed by data) →

All Jeff Su videos · Jeff Su profile · Jeff Su (official site)