DeepSeek 671B params on Mac Studio

Alex Ziskind · 195.1K views · 3.7K likes Short

Analysis Summary

20% Minimal Influence

mildmoderatesevere

“Be aware that the 'ridiculous' quantization mentioned significantly reduces model accuracy, which is glossed over in favor of showing raw speed and memory usage.”

Ask yourself: “Is this structured to help me understand something, or to keep me watching?”

Transparency Transparent

Human Detected

95%

Signals

The transcript exhibits clear markers of authentic human speech, including conversational tangents and unpolished sentence structures. The content is a first-person demonstration of hardware performance, which aligns with the creator's established human-centric production style.

Speech Patterns Natural disfluencies, filler words ('uh'), and informal phrasing ('what is that 512 yeah') characteristic of spontaneous human speech.

Personal Context The speaker references their specific hardware setup (Mac Studio) and real-time reactions to the output quality.

Channel History Alex Ziskind is a known tech personality who consistently produces original, human-led hardware and software reviews.

Worth Noting

Positive elements

This video provides a real-world data point on the memory requirements and token generation speeds for the DeepSeek R1 model on high-spec consumer hardware.

Be Aware

Cautionary elements

The video minimizes the trade-off between quantization (shrinking the model) and actual intelligence/accuracy, potentially misleading viewers about the utility of the output.

Influence Dimensions

How are these scored?

About this analysis

Knowing about these techniques makes them visible, not powerless. The ones that work best on you are the ones that match beliefs you already hold.

This analysis is a tool for your own thinking — what you do with it is up to you.

Analyzed March 17, 2026 at 12:04 UTC Model google/gemini-3-flash-preview-20251217 Prompt Pack bouncer_influence_analyzer 2026-03-15b App Version 0.1.0

More on This Topic

Related content covering similar topics.

GLM 5 on cluster of Mac Studios

Alex Ziskind

Minimal Transparent

mac studio apple silicon

$5,000 Mac Studio vs. $4,000 Gaming PC

Joe Urbinati

Minimal Transparent

mac studio apple silicon

Mac Studio vs. Nvidia: Running Large Models Locally!

Heavy Metal Cloud

Minimal Transparent

mac studio apple silicon

Memory Wars! Apple vs Ryzen - Is Unified Memory Faster than Shared GPU Memory?

Dave's Garage

Low Mostly Transparent

apple silicon unified memory

Your Mac Has Hidden VRAM… Here's How to Unlock It

Alex Ziskind

Low Mostly Transparent

apple silicon unified memory

Transcript

this is the first time I'm running the full 671 billion parameter deep seek R1 quantized of course but I'm getting [Music] 14.89% gigs of RAM what is that 512 yeah and I'm only using 175 for this full thing because this Quan ties down to uh something like one or [Music] two something ridiculous but I asked it for a JavaScript function and it gave me at least JavaScript I don't know if it's right but it looks like JavaScript so it's not terrible