😺 Gemini's big move

PLUS: Jailbreaking a robot is HOW easy?!

Welcome, humans.

Check out this “amateur” footage of close encounters with animals in a kayak:

The last one is still an animal, it’s just… not the kind you want a close encounter with…

These clips are still a little dreamy, but it’s WILD how realistic AI videos are getting.

Some of these shots (like the ones with the Pelican and Manatee) FEEL real, especially because they look like they could’ve been taken by a smartphone.

And if us young’uns here at The Neuron are having problems? How hard will it be for our parents and grandparents to tell what’s real and what’s AI?

Speaking of grandparents…

It’s… surprisingly wholesome?

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

We break down what everybody wants in AI right now.
Meta partnered with Blumhouse to test its Movie Gen video generator.
NotebookLM announced you can now “customize” your AI podcasts.
Sam Altman's Worldcoin launched a bunch of features.

Advertise in The Neuron here.

What does everybody in AI want right now?

If you want to understand where the future of AI is going, you gotta look at what the major players are doing—or more importantly, what they need.

Anthropic needs more money. OpenAI needs more power. And Google? Apparently it needs to switch up its org structure.

According to a new blog post, Google moved the company’s Gemini chatbot out from under its search and ads unit into its DeepMind AI research team.

There’s just one problem: developers don’t like working with Gemini. That’s partly because of the entrenched market share of OpenAI—its been around longer, and it has the users (about 3B monthly views, now that school’s back in session).

It’s also partly because using OpenAI’s devs tools have less friction to set up. It’s so easy to connect to OpenAI’s API (the way apps connect to other apps), someone was able to put ChatGPT on a TI-84 graphing calculator:

I Made The Ultimate Cheating Device

Don’t forget—the government also wants something with AI. And it’s not just to regulate it. Just look at yesterday’s press release from the U.S. Treasury Department.

The Treasury used AI to recover over $4B in potential fraud last year. That’s:

$2.5B from preventing high risk transactions.
$1B from foiled check-fraud schemes.
$500M from “enhanced ‘risk-based’” screening.

Sounds like a lot of money, right? Juniper Research says losses from online payment fraud could hit $362B by 2028.

FROM OUR PARTNERS

Tired of Paying for Multiple Creative AI Tools? Superstudio Has You Covered.

Superstudio is your all-in-one solution for seamless image and video creation, powered by an infinite creative canvas (think supercharged Figma).

This new tool empowers you to push your creative boundaries with top-tier models.

We’re talking:

Luma Lab’s Dream Machine.
Black Forest Labs’ Flux.
Plus Kaiber’s creative AI tools…

Get everything you need in one place …WITHOUT the need for multiple subscriptions.

With Superstudio’s intuitive interface, you can:

Combine AI generated visuals.
Refine them to perfection.
Turn your vision into reality—faster than ever before.

Try Superstudio for free today!

Around the Horn.

Meta announced it’s teaming up with Blumhouse (horror movie producers) to test out its new Movie Gen video generator.
NotebookLM now lets you guide the conversation of its podcast-style audio overviews via a new “Customize” prompt (more about this + demos here).
Perplexity now lets you search your internal files alongside web content (demo video).
Sam Altman's crypto startup Worldcoin rebranded to “World,” unveiled a new iris-scanning Orb + plans to roll out verification at coffee shops or at home, introduced Deep Face for combating deepfakes, launched a beta World ID credential, and announced a new blockchain (World Chain) + app.

Treats To Try.

Announcement thread here. Also, if you have access to Canvas, you can now track changes with the “Show changes” icon.

Kagi is a search engine that now automatically downranks AI-generated content. For example, here’s a search for baby peacocks… we caught at least 3 AI generated images (pretty similar to Google, if not a little better?) BONUS: here’s a ublacklist feed of sites who push out too much AI content.
Feta helps you run efficient meetings by auto-compiling updates, managing tasks, and generating smart summaries (in beta atm).
Kick automates your bookkeeping tasks, like categorizing transactions and finding deductions (you can sign up + apply for the free tier to start).
Convo interviews your users in their language, giving you instant insights from hundreds of conversations at once.
You can now chat with Meta's Llama AI directly inside your apps using one of the fastest chat services (SambaNova + Gradio) with only a few lines of code to set up. Try it here.
Hallo2 is an open-source avatar video model that turns your photo into a talking 4K video that matches your voice for up to an hour (code here).
Dottxt’s open-source project Outlines helps your AI write in formats your software can actually use (raised $11.9M).

See our top 51 AI Tools for Business here!

FROM OUR PARTNERS

Learn GenAI app development from NVIDIA, Databricks, HP, and more

Do you know how NVIDIA, Databricks, Twilio, HP, and ServiceNow get their GenAI apps into production? Learn their best practices at GenAI Productionize 2.0, including:

How to design a GenAI stack for enterprise scale
Techniques for AI governance, evaluation, and observability
Proven strategies for getting GenAI apps into production

Free Registration!

Intelligent Insights

This is an absolutely WILD thread.

Researchers at the University of Oxford trained an AI to act as a mediator, and found it did a better job of finding common ground and “left groups less divided.”
Anduril’s new military drones include a kamikaze mode (video here), and now defense tech founders are publicly sparring over whether AI should make lethal decisions in combat. Btw, Reddit shared that the DoD just changed its policy to authorize unmanned lethal force when assisting “civilian authorities.”
Check out this comprehensive overview of how GPT-o1’s reasoning works across two categories: prompt-based and learning-based approaches. Related: The top 7 AI papers trending on HuggingFace from October 17th,
Pietro Schirano fine-tuned an AI model on Naderi Yeganeh’s “mathematical art”, which uses math to make images…it’s pretty awesome.