😺 YouTube's Netflix killer

PLUS: Insights from under the hood of o1...

Welcome, humans.

You know how when you use o1, you can hit the down arrow and see a live feed of o1’s “reasoning” in action? Well, o1 has been caught sharing some wild thoughts in those feeds:

How about this one, where o1 had to “clear its head” before responding…

As it turns out, that “chain of thought” feed? It’s totally made up. It’s actually just a “model generated summary.” Why?

To keep how o1 actually “thinks” hidden from view. OpenAI is real serious about keeping o1 a secret… some users have even been threatened with ban warnings for asking too many questions about o1 reasoning.

Even still, o1 users on Reddit are loving the end result. And the LMSYS leaderboard just put o1 waaay ahead of the competition. Guess all that “head clearing” is working!

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

YouTube will add DeepMind's AI video generator to Shorts.
T-Mobile will work with OpenAI for customer service, NVIDIA for 6G.
Alibaba released 3 new Qwen 2.5 models for language, coding and math.
Microsoft launched an impressive new small model called GRIN MOE.

Advertise in The Neuron here.

YouTube is going all-in on Gen-AI.

It was only a matter of time, and that time has come: YouTube is getting an AI video generator.

Everybody knows Google DeepMind is a big deal in AI. The team’s prowess is more or less why Elon and Sam Altman started OpenAI—so Google wouldn’t control all the AI.

Now, Google DeepMind’s Veo, an AI that generates artificial videos, will be integrated into YouTube Shorts so you can create + post 6-second clips directly on the platform.

When this gets rolled out “later this year”, the only thing stopping you from making a viral video about a cat playing the saxophone is taking the time to write the prompt into Veo and hit post.

If you thought your Shorts feed was already full of “AI slop”, you ain’t seen nothing yet!

^ Luckily, YouTube thought of that, which is why all Veo vids will be watermarked with DeepMind’s SynthID + an AI warning label.

Veo demo in action

YouTube has a host of other AI features, too:

Inspiration: Generates suggestions + content ideas for creators.
Dream Track: generates instrumental tracks for Shorts.
Music AI Sandbox: Creates musical “loops” for content.
Auto-Dubbing w/ Expressive Speech: Improved language translation that imitates the acoustics, intonation, and pitch of the original video.

YouTube also announced some cool, non-AI related changes, like the ability to designate “Seasons” and “Episodes” for serialized content.

Our take: It’s no coincidence YouTube announced you can serialize (publish in installments) your content the same day it announced text to video on Shorts. You’ll soon have everything you need to launch your own media franchise directly on the platform—no Hollywood required!

Speaking of Hollywood… the studio Lionsgate is working with Runway to train its new model on Lionsgate movies (Hunger Games, John Wick, etc)—But only Lionsgate can use it.

The end game here is obvious: YouTube becoming Netflix. Except unlike Netflix, anyone will be able to make the content. Which will maybe make it better than Netflix? Or maybe worse than Netflix… TBH, it’s hard to get much worse than The I-Land…

FROM OUR PARTNERS

Crush your sales targets with automated follow-ups from Attention.

In today's tough economy, the key to sales success is lightning-fast follow-up. Attention's AI assistants are your secret weapon for turning conversations into conversions.

Imagine having a virtual team that:

Crafts tailored follow-up emails instantly after each call.
Provides real-time coaching to perfect your pitch.
Automates CRM updates, saving you 20-30 minutes per call.

Attention delivers this capability to 100+ top sales teams on the daily.

“Attention isn't just a tool; it's like having an army of virtual assistants at your disposal.” —Yash Tekriwal, Head of GenAI @ Clay

Don't let another hot lead cool down. You need Attention right now.

Around the Horn.

While good to know, LinkedIn already scraped all user data for training.

T-Mobile announced a multi-year partnership with OpenAI to build an agent-powered customer service platform called IntentCX, and another partnership with NVIDIA to use AI for managing network loads in anticipation of “6G.”
Qwen 2.5, the latest open-source family of models from Alibaba Cloud is out now, in three different varieties: 2.5, 2.5-Coder, and 2.5-Math (models here).
Microsoft released GRIN MOE (GRadient-Informed MoE), a 6.6B small model that’s good at coding and math (try it here, code here).

Treats To Try.

*DIVACS is the AI wingman for B2B success—its platform builds, manages, and showcases your value proposition to all your prospects, customers, and vendors. Don't leave your deals to chance—build your business case with DIVACS today.
Bolna provides 24/7 virtual receptionists to answer all your inbound calls and manage your front desk operations.
Death Clock predicts your life expectancy based on your habits and provides suggestions to help you live longer (interesting review).
CogVIdeoX released a new image to video model that lets you use an image as a background with a prompt (try it here, model here).
Cubby is a private workspace that lets you store anything (articles, PDFs, YouTube vids, podcasts, Tweets, ebooks, etc) and annotate with AI.
Zivy organizes Slack and email messages into prioritized stacks, reducing notification overload so you can focus on important tasks.
Stytch protects your apps from bots and fraud by identifying unique devices without using CAPTCHAs.