Meta's AI runs locally on Mac Studio

SPEAKER_1 (0:01)

When you need to build up your team to handle the growing chaos at work, use Indeed Sponsored Jobs. It gives your job post the boost it needs to be seen and helps reach people with the right skills, certifications and more. Spend less time searching and more time actually interviewing candidates who check all your boxes. Listeners of this show will get a $75 sponsored job credit at indeed.com/podcast. That's indeed.com/podcast. Terms and conditions apply. Need a hiring hero? This is a job for Indeed Sponsored Jobs.

SPEAKER_2 (0:31)

Low energy, unfocused, foggy? You might be dehydrated, whether it's hot yoga, over 100 degree weather or too much sun.

Gator Lite with a specialized blend of five electrolytes was scientifically designed to help move fluids into the body faster for rapid rehydration. Shop now at retailers nationwide. Gator Lite hydrates faster than water. Is it in you? Symptoms noted are signs of mild to moderate dehydration. Consult a health care professional if symptoms persist.

SPEAKER_3 (1:02)

Now at Lowe's. Faster Quotes start at the Lowe's Pro Desk. Got a material list handwritten on a sticky note or saved as a photo? Perfect. Bring it to us and get a quote in minutes. And if you don't see what you need on the shelf, we'll help you get it. You can access thousands of products beyond what's available in store or on lowes.com, right from the Lowe's Pro Desk.

Build quotes faster and source the materials you need to keep your jobs moving. Just like that, at Lowe's.

SPEAKER_4 (1:30)

Apple and Meta have formed a partnership to natively run Meta's 816 billion parameter, Muse Spark Open AI model on Apple's high-end Mac Studio computers. We're essentially looking at a hardware and software integration that places frontier scale artificial intelligence directly onto local machines.

SPEAKER_5 (1:46)

Which is, I mean, wildly ironic considering the history between these two companies. Tim Cook and Mark Zuckerberg have spent the better part of a decade publicly at war over everything from user privacy to fundamental business models. Seeing them suddenly align to optimize the largest open-weight AI model in the world for Apple's silicon is something very few people saw coming.

SPEAKER_4 (2:09)

Yeah, the market demand is simply overriding that historical friction.

Businesses and developers are hitting a wall when it comes to relying entirely on centralized cloud data centers. They want the capability of a vast multimodal intelligence sitting quietly under their own desk, fully contained within hardware they physically own.

SPEAKER_5 (2:25)

That shift in consumer behavior force is an overarching question that we really need to keep in mind as we look at this integration.

Because if a company can now run an intelligence that previously required a warehouse size server farm on a computer sitting in their office, what happens to the multi-billion dollar cloud computing industry? I mean, the entire enterprise tech economy relies on renting computational power by the hour.

SPEAKER_4 (2:47)

Right. To understand why cloud providers might be sweating it, we have to look at the actual specs of Muse Spark Open. It is an 816 billion parameter model, but it operates with 42 billion active parameters at any given moment.

Plus, it has a 1 million token context window, and it is natively multimodal, meaning it processes text, images, video, and audio directly.

SPEAKER_5 (3:12)

That distinction between total parameters and active parameters is like the entire secret to making this run locally. You can think of a parameter as a single learned connection in the artificial brain.

Having 816 billion of them means the model possesses a vast repository of logic, facts, and patterns.

SPEAKER_4 (3:33)

But if it had to activate all 816 billion connections, every single time you asked it a question, the raw electrical draw would just melt the hardware.

SPEAKER_5 (3:41)

Exactly. So it relies on a routing mechanism, specifically a sparse mixture of experts architecture. Instead of spinning up the entire model for a single prompt, it only wakes up the specific 42 billion parameters that actually apply to your request.

If you ask it to analyze a complex Python script, it routes the tokens directly to the parameters that were heavily trained on programming logic.

SPEAKER_4 (4:05)

It's kind of like walking into a massive research hospital, right? There are thousands of specialists in the building, but if you walk in with a sprained ankle, the triage system doesn't call a neurologist and an oncologist into your room. It just calls the orthopedic doctor. The model is doing the exact same thing with its data.

SPEAKER_5 (4:21)

Right. It leaves the audio processing and image generation parameters completely dormant while it handles your code block. That keeps the power draw and the thermal output manageable for a desktop machine.