The $3T AI Opportunity Everyone Missed
4 points by beiconic 8 hours ago | 4 comments
  • beiconic 8 hours ago |
    Author here. This analysis explores how programmatic audio/video advertising could drive unprecedented GPU demand. Key technical points:

    * Current compute requirements for video generation: 10x

    * Audio personalization: 5x

    * Real-time optimization: 3x

    * Intent processing: 2x

    Happy to discuss technical details, infrastructure scaling challenges, or specific implementation questions.

  • JohnFen 8 hours ago |
    > The real opportunity isn't in API fees or enterprise AI. It's in something far bigger: AI-generated, personalized audio and video advertising. Think programmatic advertising, but orders of magnitude larger.

    That this will happen is one of my biggest fears about genAI, and one of the reasons why I'm hesitant to use it as a consumer. It promises to amplify everything that makes the ad industry so objectionable today.

    • beiconic 8 hours ago |
      Valid concern about the potential downsides of hyper-personalized advertising. A few counterpoints to consider:

      1. Relevance: AI-generated ads, when done right, could actually make advertising less objectionable by making it more relevant and useful to consumers. Less spam, more signal.

      2. Opt-in models: The industry is moving towards more privacy-centric, opt-in approaches. AI could enable better alignment between consumer preferences and ad content.

      3. Regulation: As with any powerful new technology, robust regulation will be key to aligning AI advertising with consumer protection. GDPR-style data protections, disclosure requirements, etc.

      4. Benefits to consumers: AI ads could fund better free products/services, as we saw with search and social. Trade-offs exist.

      From an infrastructure perspective, I believe we have the technical foundations to do this in a privacy-preserving way:

      1. AI ad platforms could process data on-device vs. centralized servers, enabling better privacy.

      2. Differential privacy techniques can be used to process ad targeting data without directly accessing individual info.

      3. Federated learning could allow ad targeting models to be trained without raw data ever leaving user devices.

      So while the scale of data processing will be immense, I believe it's a solvable challenge with the right R&D focus and policy frameworks. Not saying it'll be perfect, but I believe aligning incentives between advertisers and consumers is possible with the right approach. What safeguards would you want to see?

      • JohnFen an hour ago |
        Well, relevance doesn't actually matter to me. Actually, if I'm to see ads, I find it less objectionable if they aren't relevant to me personally because relevance indicates that someone is spying on me.

        Regulation is desperately needed, but I seriously doubt that anything meaningful will happen in the foreseeable future in my part of the US. Nothing has happened yet, anyway, despite lots of talk about it. I mean, who knows? It's possible. However, I don't expect it.

        The benefits to consumers part is true, but it needs to be opt in (as you mentioned). I have no issue with tracking people who have given informed consent to be tracked. But again, I have no expectation that's actually going to happen. I put a fair bit of effort to avoid being tracked, but advertisers put even more effort into subverting my wishes. I should be able to have them leave me alone.

        I find the three points in your list of technical foundations to offer some improvement, but they aren't anything like a solution. They're more like reducing the bleeding. All of them would still be unacceptable (to me) in the absence of my informed consent.

        > What safeguards would you want to see?

        To me, the underlying point is that unless I've consented to having information about me, my machines, or my use of my machines tracked, I'm being spied on and object to it strongly. It's hard for me to think of what technological safeguards would help here. I can think of legislative ones, though. For example, DNT could have gone a long way if it weren't voluntary.