😺 Google builds its own JARVIS

PLUS: Redditors trick Google with fake reviews

Welcome, humans.

In honor of Halloween this week, here’s a video of Gordon Ramsay taking on some of Hollywood’s creepiest horror characters, and TBH, we don’t know who is scarier:

The score is hardcore—so turn your volume down before you hit play!

These Gordon Ramsay AI videos are so popular, The Information ran a report on them. And if The Information runs it, you know it has legs. The report was all about Chinese video generators, like Hailuo who makes MiniMax, Kling, PixVerse, and Vidu—and how these models have an early lead on those in the U.S., like Runway.

Not to be outdone, Runway released Act One this weekend, which lets you turn any performance (like you recording yourself in a mirror) into an animated character or photorealistic avatar.

Creators have already used Act One to create scenes from 1960s gothic horror movies, animated zombie shorts, funny monologues, and whatever the heck this is (plus more demos here).

Now, this short film about a Capybara visiting a ramen restaurant is NOT built with Act One, but it’s adorable. A Capybara’s big day out in the city? Eating ramen?! The only thing missing is the Capybara song!

Here’s what you need to know about AI today:

Google’s got an AI called Jarvis that can use your computer.
Apple will release M4-powered Macs for AI features this week.
Humane cut AI Pin price, shifted to OS licensing.
Redditors tried to trick Google AI with fake steakhouse reviews.

Advertise in The Neuron here.

Google is building an agent called Jarvis that can use your computer like Claude can.

Remember Ask Jeeves? If you never used the internet in the late 90s or early 00s, it was like Google before Google, only way worse.

Well, apparently Google is now building “Jarvis”, an AI agent that can use your computer (like Claude now can), set to be previewed as soon as December. Oh, and before you ask, yes, Jarvis is based on J.A.R.V.I.S, Tony Stark’s AI wingman.

Here’s how Jarvis will work:

Jarvis will be consumer-facing Google Chrome tool that takes screenshots of your screen and interpreting them.
It should be able to automate web based tasks by taking actions like clicking or typing.
Jarvis will also be powered by Gemini 2.0, which should also come out in December.

So pretty soon you’ll be able to “Ask Jarvis” fill out your expense reports, do your homework, or even look up what Ask Jeeves is for you.

This is cool because of Google’s scale, but Google’s in last place here:

OpenAI has been developing AI for computer use since February at least.
Anthropic just released computer use last week.
Microsoft envisions its new AI agents as “the new application” which will interface with your computer (and Apple’s AI will eventually work like that, too).

Here’s why everyone wants AI to use your computer: Computer use is basically what’s needed to enable agents. So it’s a very important breakthrough.

For example, Microsoft and Salesforce offer “agents” that can take actions for you right now, but these require detailed instructions to set up, and aren’t yet intuitive enough to be useful for most people.

With a computer use model, like what OpenAI, Anthropic, and Google is building, you can streamline this onboarding by a lot.

Computer use also comes with some SERIOUS security concerns, though. Watch this video of a programmer who tested how easy it is to trick Claude into downloading malware onto a computer with prompt injections:

Claude Computer Use: The ZombAIs are coming! From Prompt Injection to Command & Control.

First, he tried a few sophisticated techniques to get Claude to download the malware. When those didn’t work, he tried something much simpler: just straight up ask Claude to download and launch the virus (labeled something else, of course)—and it worked.

Here’s the takeaway: This is just one technique. Nefarious actors will come up with plenty more. Sure, computer-controlling AI assistants could be amazing—but maybe let's make sure they can't be tricked into downloading malware?

FROM OUR PARTNERS

We tried Attention, and here’s what we think.

We've been testing Attention for our sales + internal calls lately, and wow—we’re actually obsessed. Here’s how it works:

Attention listens to your calls in real-time, and takes diligent notes on topics you request.
How? You identify the key data from calls you want to follow up on, and Attention analyzes the call to find the relevant info.
Then, once it’s done, you can go back and watch the relevant sections of your call to cross-reference the notes.

So why do we love it? No more scrambling to remember key talking points or frantically taking notes mid-call. We can be in the moment without worrying about jotting down every relevant tidbit. Be present, as the gurus say.

Attention can do a whole lot more than that, too.

We highly recommend you book a demo and try it out for yourself here.

Treats To Try.

*Learn GenAI tips from NVIDIA, Databricks, Twilio, and more. Tomorrow 10,000+ AI professionals will learn how NVIDIA, Databricks, Twilio, and more get their GenAI apps into production!
Don't miss GenAI Productionize 2.0 for top best practices, including:
- How to design an enterprise GenAI stack.
- Techniques for AI governance, evaluation, and observability.
- Proven strategies for getting GenAI apps into production.
Last Chance to Register.
NotebookLlama from Meta is an open-source version of Google’s viral NotebookLM feature which generates a podcast from any text (demo here).
Assembled helps you manage your customer support teams better by showing staffing needs, agent performance, and handling automation in one dashboard.
Loomos transforms screen recordings into polished videos by cleaning up speech, adding backgrounds, and offering translations in 20+ languages.
Steiner is a series of open-source reasoning models that attempts to work like OpenAI’s o1 model—read more about it here and here.
ArcadeAI lets you design custom jewelry and gets skilled artisans to handcraft it for you. (read more here).
PricingMaker helps you price your products by analyzing the cost to produce it, seller fees, and your competition (free to try, we priced a product in < 1 min).
Speaking of agents, Phidata helps you build and manage teams of AI agents that can search the web, analyze data, and solve problems while keeping a memory of your conversations.

See our top 51 AI Tools for Business here!

*This is sponsored content. Advertise in The Neuron here.

FROM OUR PARTNERS

OpenAI case study: Maven AGI AI Agents solve 93% of customer support questions autonomously

Maven AGI partners with TripAdvisor, HubSpot, Clickup, and more to improve customer experience. In a recent case study featured on the OpenAI website, Maven AGI helped answer 93% of customer inquiries, improved agent productivity by 30%, and boosted revenue retention.

Learn more about Maven AGI or get a personalized demo here.

Around the Horn.

Apple launches a new line-up of Macs + Macbooks this week (starting today) that use its new M4 chip, which is meant to handle its upcoming AI features.
Humane, the trouble maker of the AI Pin, dropped the price to $499 and shifted its strategy to license its AI operating system (CosmOS) to other companies.
About 45% of Smartphone owners say they won’t pay a monthly subscription for AI, 34% have privacy concerns, and 25% don’t find AI helpful—instead, they want more battery life, more storage, and better camera features.
Redditors have begun to write fake reviews to trick Google’s AI Overview into recommending the chain “Angus Steakhouse” in London in order to drive long lines of tourists away from their favorite local steakhouses.

What gives it away that this is AI?

Your job is to tell us what about the video below gives it away as being AI. Click the link, watch the vid, and then answer the poll with your best take!

This one really had us convinced until that one part… when you watch, you’ll know what we’re talking about.

What gives it away that this video is AI?

Write in.

Write in, pt. 2

Here’s what #Neuron readers had to say about last week’s video:

M.V: “The hand movements are too erratic, some are blurry, some are moving in weird directions. But when you don't see the hands, it's really hard to say that's it's not a real video.”
C.B: “The back of the blue floral dress is white, the eye glasses are a bit modern, the birthday cake (back then it would be frosted completely), and everybody smoked back then so an ashtray at each table. Also the lady in the peach colored tank at the end, it looks like her finger placement is off.”
D.T: “As always - AI has yet to solve the botox problem. There are hundreds of muscles in the face but these behave (although much better than it used to be) as if none of those muscles exist. So it feels oddly mechanical. And the colors are off. Whoever mimicked the films/videotape of 50 years ago needs some color management training.”