Avocados aren’t vegetables. They’re berries. I know, sounds weird.

Botanically, they count as a single-seeded berry (????), which means guacamole is technically fruit-based behaviour.

I won’t look at it the same anymore.

What’s in store:

Read Time: 5 minutes

META

WhatsApp’s AI just got incognito mode

WhatsApp is rolling out a new private mode for Meta AI chats, where neither WhatsApp nor Meta can read or store the conversation.

The feature works like an AI “incognito” mode. When it is switched on, the user’s messages and the chatbot’s replies are not monitored, and the chat history disappears from the user’s side, too.

WhatsApp head Will Cathcart said people want to ask AI about sensitive topics, such as health, relationships and money, without feeling like that information is being saved.

Meta CEO Mark Zuckerberg called it the first major AI product with no chat logs stored on servers. 

Cathcart said it is not the same as WhatsApp’s usual end-to-end encryption, but is meant to offer a similar level of privacy.

The update comes after Meta AI was added to WhatsApp last year, which annoyed some users because they could not fully turn it off. 

In May 2025, Zuckerberg said Meta AI had reached one billion users across Meta’s apps.

But the feature also raises concerns. Cyber security expert Prof Alan Woodward said the privacy risk to WhatsApp’s wider security system seems low, but disappearing chats could make it harder to investigate problems if something goes wrong.

The worry is accountability. If AI conversations cannot be recovered by users or Meta, it may be difficult to prove whether chatbot responses caused harm.

The main points:

Privacy with a paper-trail problem

Cathcart said the feature will start with text only, not images.

He also said Meta AI’s safety systems will be cautious when handling requests that may be harmful or illegal.

For now, WhatsApp users can only use Meta’s own AI inside the app, after the company blocked other chatbots from the platform.

Meta is also spending heavily on AI infrastructure, with expected costs of around $145bn in 2026. 

Investors are watching to see whether that spending leads to stronger growth across Meta’s apps, ads and commerce tools.

Meta, write me up a plan for world domination. - MV

Vote for live results + see results and opinions from yesterday at the bottom of the email.

AI CODING

OpenAI Codex vs Claude Code: We Built the Same App Twice

Codex builds apps in 18 seconds… Claude Code works with zero setup… but which one should you actually use? 

Matt Wolfe tested both tools head-to-head on The Next Wave podcast and discovered they're not competing — they're complementary. This playbook shows you exactly when to use which. 

From the AI coding showdown: 

Stop choosing. Start using both.

Number Crunch

An item costs $80. With 2% inflation for 2 years, what's the price now?

Find the answer at the bottom of the email!

Watch This

Time feels simple until physics gets involved. In this video, Sean Carroll unpacks why time may be far stranger than our everyday experience suggests, including one of physics’ biggest unanswered questions: why it seems to move in one direction at all.

AI EDUCATION

AI has made cheating harder to ignore

AI has made cheating harder to ignore.

Princeton University is bringing back supervised exams for the first time in 133 years, after concerns that AI and phones are making cheating easier.

The university’s honour code has allowed students to take exams without a professor or proctor in the room since 1893.

But faculty have now voted to require supervision for all in-person exams from this summer.

Princeton says students and staff raised concerns that cheating has become harder to spot. 

AI tools, laptops and phones have made it easier for students to switch windows, look things up or use devices during exams.

In brief:

The honour code gets a chaperone

Students will still have to pledge that they have not broken the rules. 

The main change is that instructors will now be in the room as witnesses.

If they see a possible violation, they can report it to Princeton’s student-run honour committee.

The move also reflects another issue: students are less willing to report cheating themselves. Some worry about being shamed, bullied or doxxed online by classmates.

A student newspaper survey found that nearly 30% of Princeton seniors said they had cheated on an assignment or exam. 

Almost 45% said they knew about an honour code violation but did not report it.

The update shows how AI is pushing universities to rethink old exam systems. Some are going back to handwritten exams, while others are trying AI detection tools.

Princeton lasted 133 years on vibes and integrity, then someone opened another tab. LOL.- MG

Mindstream Picks

Health: NHS-approved treatments for spinal muscular atrophy are transforming survival and quality of life for children with the condition.

Gaming: Demon Slayer: Infinity Castle has made history as the first winner of Japan’s new Cinema Awards, cementing its record-breaking global success.

Movies: Laika has released the first trailer for its star-studded stop-motion fantasy Wildwood, teasing an epic woodland adventure ahead of its October release.

Don’t Miss: Unitree has unveiled the GD01, a $650,000 “production-ready” mech suit that can switch between bipedal and quadrupedal movement. While its real-world use is unclear, the launch signals how experimental robotics is pushing further into sci-fi territory.

AI Art

Image of the Day

Daily Image Prompt

Submit your artwork to Mindstream →

Yesterday’s Poll

“Is SoftBank’s OpenAI bet a smart long-term play or too much risk in one basket?”

It’s a smart AI power move - 27%

It feels a little too concentrated - 73% ✅

Your Views:

“Even if AI may not a bubble, it can still burst!!” - info@ja

“It's a sign that Softbank is running out of good investment ideas.” - vcot

“All the AI deals and circular financing are a house of cards just waiting for the first domino to hit!” - rgra

Submit your opinions in our polls to be featured!