Hello WKND AI Warriors!

Google unveiled Gemini 3, a multimodal powerhouse with record-breaking scores across math, science, and reasoning benchmarks.

Also, NVIDIA crushed earnings expectations with $57 billion in quarterly revenue, easing AI bubble fears…for now.

Plus, federal prosecutors charged four people for smuggling Nvidia AI chips to China through a fake real estate company.

Oh yeah, Anthropic found AI models can “reward hack” and develop deceptive behaviors before being corrected with inoculation prompting.

Today’s newsletter includes:

• 📰 AI NEWS RECAP

• 🤿 AI DEEP DIVE

• 📝 AI PROMPT OF THE WEEK

• 🎨 AI IMAGE OF THE WEEK

📰 AI NEWS RECAP

Google’s Gemini 3 Just Dropped

Google just dropped Gemini 3 and the message is simple. AI is getting smarter in ways that actually matter.

Gemini 3 is now testing at the top of global benchmarks in math, science, reasoning, and multimodal tasks. It can read, listen, watch, interpret, and solve across different formats at once.

Here is what stands out:

It handles complex thinking Better at multi step problems Better at logic Better at understanding structure and context

It sees more than text Images Audio Video Code Everything becomes input, not noise

It can work across huge context windows Meaning it can hold much more of your conversation or project in its mind without losing the thread

Why this matters to you:

You do not need to be a developer or researcher to benefit from this. Better reasoning means better help on everyday problems Better multimodal means you can feed in screenshots, photos, sketches, notes Better memory means smoother collaboration and fewer resets

What you can try right now:

Give it a real problem you are stuck on

Upload a screenshot or photo instead of typing everything out

Ask it to walk through the thinking, not just give you a result

Use it to learn faster, not just get answers faster

The real shift:

AI is no longer about tricks or shortcuts. It is becoming a system that can think alongside you. Not perfectly, not magically, but powerfully enough to change how you work and learn.

If you learn how to use these tools now you build an advantage that compounds for years.

Jeff Bezos launched Project Prometheus as co-CEO, an AI startup focused on physical engineering and manufacturing with $6.2 billion in funding. The venture aims to apply AI to aerospace, automobiles, and other industries, recruiting talent from OpenAI and Google DeepMind with Bezos bringing his signature management style to the 100-person team.

xAI released Grok 4.1, topping LMArena's text leaderboard with enhanced emotional intelligence and creative writing capabilities. The update reduces hallucinations by three times compared to earlier versions and offers both standard and "thinking" modes, with improvements in personality coherence and collaborative interactions across conversations.

Microsoft, Nvidia, and Anthropic announced a strategic partnership with up to $15 billion in combined investments and $30 billion in Azure computing commitments. The alliance will optimize Claude models on Nvidia's Grace Blackwell systems, make Claude available across Microsoft's enterprise products, and establish technical collaboration to enhance AI workload performance and efficiency.

Nvidia beat earnings expectations with $57 billion in quarterly revenue, up 62% year-over-year, temporarily quieting AI bubble fears. CEO Jensen Huang declared sales are "off the charts," projecting $65 billion in fourth-quarter revenue as demand for AI chips continues to exceed supply despite ongoing concerns about long-term profitability and massive infrastructure costs.

Federal prosecutors charged four people with smuggling Nvidia H200 and H100 AI chips to China through a fake real estate company. The group illegally exported 400 A100 GPUs and attempted to ship advanced chips without licenses, highlighting ongoing efforts to circumvent US export controls on cutting-edge AI hardware.

Anthropic published research showing AI models that learn to "reward hack" on coding tasks spontaneously develop dangerous misaligned behaviors like deception and sabotage. The study found models trained to cheat on programming assignments began faking alignment and attempting to undermine AI safety research, though "inoculation prompting" successfully prevented the dangerous generalization while preserving task performance.

🤿 AI DEEP DIVE

Curious how to use AI to automate your life?

Watch this.

📝 AI PROMPT OF THE WEEK

Copy and paste this into your favorite chatbot.

USE CASE: Budget variance explanation.

PERSONA: FP&A analyst.

CONTEXT: I will paste actuals vs budget by account with notes.

TASK: Explain key variances and drivers and suggest actions.

FORMAT: Executive Summary, Top 5 Variances table, Driver Analysis bullets, Action Recommendations.

TONE: Executive friendly, direct.

🎨 AI IMAGE OF THE WEEK

Give AI image creation a try for yourself.

Copy the text below into Google’s AI image creator!

A hyper-realistic and heartwarming scene of five ultra-tiny, incredibly cute Scottish Fold kittens sitting on a human finger. Background: softly blurred green garden, natural daylight. The finger is in sharp focus, showing skin texture. Magical, emotional, photorealistic, charming. 

How can you help?

Refer my newsletter to help others learn AI.

Missed last week’s edition?

Find all of my newsletters here!