Hello and welcome to the Automated, your AI tour guide.

Google just took a hit, and Perplexity made sure it stung.

In a bold new ad, they didn’t just mock Google’s AI fail—they exposed a bigger problem lurking in AI search. And as it turns out, bad answers aren’t just Google’s issue… they’re everywhere.

Here’s what we have for you today:

AI search engines are supposed to be the future, but right now, they’re mostly just taking swings at each other.

And Perplexity? Oh, they’re not fighting fair. In-fact, they just threw a Squid Game-level punch at Google in their latest ad.

Starring Squid Game legend Lee Jung-jae, the ad shows him trapped in a room, forced to answer a series of questions to escape.

Then comes the knockout question: “How do I make cheese stick to a pizza?”—a not-so-subtle jab at Google’s infamous AI fail last year, when it confidently told users to add Elmer’s glue to their pizza sauce. Yep. It was absolute savagery.

Lee first tries a Google knockoff called “Poogle” (a hilariously subtle touch), gets garbage results, then turns to Perplexity’s AI assistant which confidently hits back with the actual correct answer: “Use fresh, low-moisture mozzarella. Don’t use glue.”

Honestly, That burn is so smooth it deserves a Michelin star.

But here’s the thing—Perplexity is throwing punches in a fight where everyone, including them, is guilty.

According to fresh research from the Columbia Journalism Review, AI search engines like Perplexity, ChatGPT Search, and Google’s Gemini are 60% of the time failing at their one job: getting answers right.

And some of them aren’t just wrong—they’re spectacularly, embarrassingly wrong.

Here’s how they stacked up in the accuracy test:

But hey, it gets even worse. Not only do these AI models mess up constantly, they do it with alarming confidence

They don’t second-guess themselves, they rarely admit when they don’t know something, and worst of all? They barely credit their sources.

Traditional search engines act as middlemen, guiding users to real sources—news sites, research papers, and actual experts.

But AI search? It just steals contents, repackages it, and presents it as fact—often without linking back to the original publisher. The study found:

This is why Google’s push to shove AI Overviews into more searches (whether users want them or not) is kind of terrifying. Imagine replacing real search results with AI-generated hallucinations that confidently feed you nonsense.

And then there’s journalism.

If AI search keeps swallowing up content without sending traffic back to publishers, news sites will struggle to survive. Less traffic means less funding, which means fewer journalists. And fewer journalists? That’s how you get an internet full of nothing but AI-generated garbage.

Honestly? No one.

Perplexity might be leading the pack, but when your accuracy rate is only 63%, that’s not exactly a win.

Meanwhile, Google is doubling down on AI Overviews, ChatGPT is trying to take over search without crediting sources, , and Grok… well, let’s just leave Grok out of this.

Well… only if you double-check everything, don’t mind a few hallucinations, and for the love of Pizza, absolutely avoid or ignore any AI-powered cooking advice.

Otherwise, It might be safer to just—dare I say it—Google things the old-fashioned way.

But hey, if nothing else, at least Perplexity is making the AI chaos entertaining.

For the full breakdown, check out the report.

Hi Automated Assembly!

This is the lecture I shared last week in Fukuoka with the launch of Nu Source, my friend Hashimoto San’s venture studio. I did this lecture in a movie theatre (a firsts!) and it was really was a combination of both the university lectures I do as well as feedback from students on what resonated (and what didn’t!).

And はじめまして to all the new Japanese subscribers into this community, especially the Nu Labbers!

I'm also excited about our new Premium newsletter coming out tomorrow. I’ve got loads of insights I gained on how to use AI into my workflows - and looking forward to sharing it with you.

Big news to announce - our YouTube show is launching today!

We’ve got two playlists: 

Season 2: Where there are great interviews like with the premier episode with Dr. Hunaid Gurji on AI and medicine, and my monologues on geopolitics in the age of AI

The other playlist is our collaboration with HeyGen, where we turned our posts into avatar presented news. So you get to consume the same great content that you love from The Automated, but with avatars and on YouTube!

We have a whopping 235 videos and over 12 hours of content in the pipeline- so check it out!

Cheers,

Tak, Editor at The Automated

Now, onto the next big piece of AI news...

Nvidia is making a jaw-dropping $2,300 in profit every single second, and it’s all thanks to the AI boom.

But instead of sitting back, it’s charging full speed ahead with its next-gen AI Superchips: Blackwell Ultra GB300 (2025) Vera Rubin (2026), and the Rubin Ultra (2027).

These absolute powerhouses are designed to keep Nvidia on top, and the numbers are just ridiculous.

The Blackwell Ultra GB300 is up first, and while it’s not a completely new architecture, it still packs a serious punch.

It delivers 20 petaflops of AI performance—same as the original Blackwell—but cranks up the memory to 288GB of HBM3e, a 50% increase over the previous 192GB.

The Blackwell Ultra DGX GB300 "Superpod" cluster isn’t playing around either, featuring 288 CPUs, 576 GPUs, 11.5 exaflops of FP4 computing, and 300TB of memory.

Then there’s the GB300 NVL72, a single-rack beast that offers 1.1 exaflops of FP4 AI compute, 20TB of HBM memory, 40TB of “fast memory,” and 130TB/sec of NVLink bandwidth.

But Nvidia didn’t dwell on Blackwell Ultra for long before introducing its real game-changer: Vera Rubin, coming in 2026.

This is where things start to get insane. Vera Rubin more than doubles Blackwell Ultra’s AI compute, pushing 50 petaflops of FP4 performance.

A full Vera Rubin rack is expected to deliver 3.3 times the performance of a comparable Blackwell Ultra setup.

And finally comes the Rubin Ultra in 2027, an even more monstrous version that effectively fuses two Vera Rubin GPUs into one.

This 100 petaflop giant comes with 1TB of memory—four times what Blackwell offers—and promises to be 14 times faster than Blackwell Ultra racks.

Demand is already through the roof.

Nvidia has shipped $11 billion worth of Blackwell GPUs, and just four buyers have already ordered 1.8 million Blackwell chips in 2025 alone.

And if you thought Nvidia was stopping here, think again—CEO Jensen Huang confirmed that after Rubin Ultra, the next AI chip will be called Feynman, set to launch in 2028.

Huang made it crystal clear: the world is going to need way more AI power than anyone expected.

While some investors started to worry that AI efficiency improvements might slow down Nvidia’s sales, Huang shut that down, saying the industry now needs 100 times more computing than we thought last year.

Nvidia is making it clear—the future of AI computing belongs to them, and they’re pushing faster, bigger, and stronger than ever. Buckle up, because this AI race is just getting started.

Click here to read more about these exciting updates.

In 2 years you will be working for AI

Or an AI will be working for you

Here's how you can future-proof yourself:

Join 1,000,000+ pros at companies like Google, Meta, and Amazon that are using AI to get ahead.

Sign up and start learning AI

The right email can turn a maybe into a yes!

Whether it’s a promo, a special offer, or a loyalty reward, your message needs to grab attention and drive action. Ready to craft emails that sell?

Here are some prompts to help you get started!

Source

This should help streamline your interactions with ChatGPT and get the results you need more efficiently.

Best of all, It's free!

Click here to get it now.

That's all we've got for you today.

Did you like today's content? We'd love to hear from you! Please share your thoughts on our content below👇

Your feedback means a lot to us and helps improve the quality of our newsletter.