Can’t waste precious time here.

The AI scene was so wild this week, you’d swear Big Tech was sleeping on corduroy pillows with the amount of headlines they were making.

We gave it the normal ‘AI news that matters’ treatment today, so let’s drive straight into what you need to know.

Here’s the gist shorties.

Now we've got AI models just checkin other AI models at the door? 

Yuuuuup.

Meta just unveiled a new "Self-Taught Evaluator" model designed to assess AI responses with minimal human input.

It utilizes a "chain of thought" technique to break down complex problems.

And Meta got super…. meta with the training. They trained the evaluator model entirely on AI-generated data.

This new evaluator model could pave the way for digital assistants capable of performing a wide range of tasks independently.

What it means: 

We're not QUITE yet to the point where AI models are making new AI models, but using AI/synthetic data is already all the rage.

Expect a significant reduction in costs and faster iteration cycles for AI companies.

(Yeah, even humans working in the AI space have AI competing for their jobs. Weirdy, right?)

But this synthetic AI data and AI models fact checking each other could also be a MASSIVE ethical minefield if left unchecked.

(See what we did there? lolz)

This model didn't make a ton of waves, but expect more of Big Tech to follow Meta's lead here.

Hot off the presses – Microsoft just announced details on its upcoming AI agents in Copilot Studio. 

As part of its “AI Tour” even in London, Microsoft showed off an AI agent example from McKinsey that handled inbound engagement leads. They also said access should be rolling out to the general public in November. 

Microsoft execs said that McKinsey’s AI agents cut lead time by 90% and cut down on admin work by 30%. 

What it means: 

Yeah, we’ve been hearing about AI agents for seemingly two years. Those early rumblings were waaaaayyyy too early, in our opinion. 

Now, AI agents that can perform actions on your behalf are going mainstream. 

With Salesforce announcing Agentforce and Microsoft rolling out autonomous agents in Copilot Studio, the AI agent game is going from AI-fueled fairytale to how we’ll all be working soon. 

Get on board y’all. 

Guess who's movin' into your Windows neighborhood? 

The ChatGPT app!

….. finally.

After OpenAI rolled out its ChatGPT desktop app to Mac users back in June, it finally opened up access for Windows users. 

ChatGPT Plus, Team, Enterprise, and Edu users get first dibs, as the app is not yet available for free users. 

What’s the advantage? 

Aside from a dedicated app interface and the UI improvements that brings, searching inside chats is finally available in the app version! And, it’s more intuitive to use ChatGPT directly integrated with your operating system. 

Buuuut... it's still missin' some party tricks. No voice support yet, and some GPT Store integrations are MIA. For now. 

We already did a thorough run-down on the ChatGPT desktop app for Mac, and most of this should still apply for Windows users. 

What it means: 

AI's movin' on up from browser tabs to your taskbar, y'all.

This is OpenAI planting their flag in your digital real estate. They're making it reaaaal comfy for users to stick with their platform. 

Early on, we didn’t think the desktop app experience would roll out to Windows users, as the train of thought was that it would stomp on Microsoft’s Copilot experience. 

And with Microsoft’s Ignite Conference less than a month away, keep it eye out for more GPT-powered offerings making their way into the OS. 

(Well, unless things get sour between the two parties. Keep reading on for that.) 

If you’ve been paying attention to Everyday AI, you’re not surprised by this report. 

(Pretty obvious, TBH.) 

Word on the street (aka Bloomberg) is that some Apple employees think they're about two years behind in the AI race.

At WWDC24, Apple flexed with "Apple Intelligence" – showin' off AI notification summaries and a Siri glow-up.

But, it’s nearly 2025 and Apple has yet to roll out Apple Intelligence to all users. And it’ll probably take at least another year or more. 

(Yikes.) 

What it means: 

Apple's AI lag could bite businesses that are all-in on their ecosystem.

Sure, Apple's pulled off some clutch comebacks before (remember Apple Maps?), but AI's movin' at warp speed, y’all. 

Sure, Apple is taking privacy WAY more seriously than other companies when it comes to integrating GenAI and LLMs into their hardware and software.

But at what cost? 

By the time Apple fully releases its Apple Intelligence, most consumers aren’t going to care. Because GenAI has already been hitting us at all angles for the past 18 months. 

Apple thinks AI is gonna reverse its slumping sales. 

We don’t see it. WAY too little. WAY too late.

Perplexity AI just dropped some new features that'll make your research game go brrrr.

They're callin' 'em Internal Knowledge Search and Spaces. 

Now you can search both the web AND your team's files at the same time. It's like having a super-smart intern who never sleeps.

Buuuut... there's always a but.

Perplexity's caught some legal heat from media outlets. Something about plagiarism and cease-and-desist orders. Rut-roh.

.

What it means: 

There’s a bigger play going on here in our opinion. 

Perplexity has already carved its path as an AI answers engine and an alternative for ‘old school’ manual search. They’ve already got Google in their crosshairs. 

With these updates? 

They’re going after ChatGPT, Claude and Gemini.

They’re hoping adding features similar to GPTs (ChatGPT), Projects (Claude) and Gems (Gemini), that many casual LLM users might drop one of those big three and instead go all-in with Perplexity. 

We still gotta go deeper on these new features. 

This isn’t scary at all. *

*Kidding. It’s actually kinda/pretty scary.

The U.S. Military is going all in on AI deepfakes according to reports from the Intercept. The U.S. Special Operations Command reportedly wants to work on deepfake technology to create realistic online personas that look like real people.

Sounds like nothing could go wrong, right? 

(Double yikes) 

The Pentagon is interested in deepfake technology, even though U.S. intelligence warns about the risks of synthetic media, especially from foreign enemies. 

Experts worry that the military’s use of deceptive technologies could harm public trust in government information, making it harder for people to tell truth from fiction.

What it means: 

Welcome to the future, where seeing ain't necessarily believing.

This raises HUGE questions about information integrity and public trust.

Businesses need to up their digital BS detectors, pronto. Invest in rock-solid authentication methods and train your team to spot digital fakes.

We’ll keep a close eye on this story in the future as it unfolds, as there’s something SUPER unsettling about the DoD saying it wants to explore deepfake tech. 

Who’s there entering the LLM cat fight? 

…..

NVIDIA? 

Yup. NVIDIA just crashed the AI model party with a new hotshot: Llama-3.1-Nemotron-70B-Instruct.

This bad boy's struttin' its stuff on Hugging Face, flexin' some impressive benchmark scores.

We're talkin' 85.0 on the Arena Hard benchmark. That's better than OpenAI's GPT-4o and Anthropic's Claude 3.5 Sonnet on these specific tests.

What it means: 

First off – this is ONE very specific benchmark. This model is in no way a general competitor to GPT-4o or Claude 3.5 Sonnet. So don’t get that twisted.

Second….Nvidia's not just the GPU king anymore – they're gunning for the AI model throne.

With unlimited funds and unlimited compute, NVIDIA is putting its smartest AI researchers to work. 

In short, we see NVIDIA masterfully playing a tricky game here. Maybe we’re just thinking about this too much, but think about this powerplay. 

By finetuning an open source model and getting similar benchmarks to the big frontier models, this puts pressure on the OpenAIs, Microsofts, Anthropics and Googles of the world to up their training game. 

And what does that require? 

NVIDIA GPUs. 

Adobe just dropped a bombshell at their MAX conference: meet Firefly, the AI video tool that could become video editor's new BFF.

This bad boy can create videos from text or images. (Once it’s fully released, that is.)

Adobe also released the V3 version of its Firefly Image editor, as well as sprinkling AI dust into about all of its Creative Cloud products.

What it means: 

The creative world's about to get flipped on its head, y'all.

Why? Two seconds. 

Yeah, AI video is cool and all. But the real magic? The new Generative Extend feature, which can extend existing video clips by up to two seconds.

That might not sound like much, but for editors, it's like finding an extra few hours in your week. 

Also, this is gonna be the first time 99% of people in the world SEE AI video. Because Adobe’s Creative Suites is how the world edits video. 

And there’s a chance, they won’t even be able to tell those two seconds are AI generated.

This Adobe drop is actually bringing AI to the masses.

Microsoft and OpenAI are doing the corporate tango, and it's getting spicy shorties. 

According to reports, the two sides are negotiating their initial blockbuster partnership. They've called in the big guns – Goldman Sachs and Morgan Stanley – to help sort out Microsoft's stake in OpenAI.

OpenAI's most recent valuation? A cool $157 billion. That's "billion" with a "b," y’all.

But here's the kicker: OpenAI's projecting a $5 billion loss this year.

Ouch.

But they're betting on that sweet, sweet future growth.

Oh, and there's this little clause that could cut Microsoft off if OpenAI cracks the AGI code. No pressure or anything.

What it means: 

This ain't your average business deal caught in negotiations.

We're watching a high-stakes poker game that could reshape the entire AI industry.

Obvi, we’ll be covering all the updates as they happen. But this is one of those negotiations that could have HUGE implications, depending on how it turns out.