What Now? With Trevor NoahIs Marques Brownlee the Most Powerful Man in Tech? | What Now? with Trevor Noah Podcast
At a glance
WHAT IT’S REALLY ABOUT
Trevor Noah and MKBHD unpack AI, influence, and tech’s decline
- Brownlee argues AI is on a steep improvement curve whose eventual plateau is unknown, creating both excitement and fear as companies overpromise outcomes.
- They discuss practical AI as a personalized assistant versus humanoid-robot fantasies, and the challenge of regulating harms like misinformation, deepfakes, and unsafe generation tools.
- Brownlee describes how AI training data often includes scraped creator content, citing datasets containing hundreds of his videos and Sora-generated “style” signals (like a familiar desk plant) as evidence.
- The conversation reframes platform algorithms as individualized “pull” systems (content retrieved for each user) rather than centralized “push,” while still holding platforms responsible for incentive structures and outcomes.
- Noah and Brownlee explore creator economics and credibility amid outrage cycles, higher production expectations, and a broader “shitification” trend of unfinished products, subscriptions, and marketing ahead of delivery.
IDEAS WORTH REMEMBERING
5 ideasAI feels scary because its improvement curve is steep and its ceiling is unclear.
Brownlee frames AI like other technologies that rapidly improve then flatten, but unlike phones or EVs, we don’t yet know where AI’s plateau is—below, at, or beyond human-level capability.
The most valuable near-term AI is an “always-on personal assistant,” not a robot replacement for humans.
He paints a best-case scenario where AI uses your phone’s context (calendar, traffic, contacts, meeting details) to proactively help, reducing friction without needing “general intelligence” embodied in humanoids.
Regulation is lagging, so companies are effectively setting the first rules themselves.
Using OpenAI’s Sora as an example, Brownlee notes “self-imposed” limits (no minors, violence, or real likenesses) are a start, but society still needs enforceable standards around harm and misinformation.
Creator content is already being absorbed into AI systems in ways that threaten originality and consent.
Brownlee says large training datasets included hundreds of his YouTube videos, enabling “in the style of” outputs; the Sora example embedding a familiar set prop suggests models learn creator-specific visual language even without explicit prompts.
Algorithms may not ‘push propaganda’ manually, but platforms still shape outcomes via personalization systems.
Brownlee recounts YouTube’s explanation: it “pulls” videos for each logged-in user based on predicted interest; Noah counters that if a platform’s system amplifies harmful content, responsibility can’t be waved away as purely user-driven.
WORDS WORTH SAVING
5 quotesNo, uh, I'm sorry, um, sir. That algorithm is based on what you like looking at.
— TikTok representative (as quoted by Trevor Noah)
I think how I describe it is the weight of the publish button is a little heavier, but it's, uh, it's a good place to be.
— Marques Brownlee
So it's going through this really, really steep improvement, and no one knows where the flat line is, which is the concerning part.
— Marques Brownlee
I've come across several data sets that have been used to train these models that include hundreds of my videos. Hundreds of my videos.
— Marques Brownlee
The star of my videos is not me. The star of my videos is the tech.
— Marques Brownlee
High quality AI-generated summary created from speaker-labeled transcript.
