32 chunks · 26 episodes
...alysis on token economics I've come across. The tech industry just assumes zero marginal costs. With tokens, the marginal cost of inference is non-trivial. Marginal costs are back, baby! The frontier labs have presumed that they'd do significa...
... is more like oil-refining than coding." We take for granted that tech has zero marginal cost, and thus has insane, otherworldly leverage. (We often mistake that hurricane force tailwind for us just being smarter than other industries, by the ...
...sandblast it with agents to make the software more robust. The sandblasting has marginal cost, but the software that is produced out the other side doesn't require marginal LLM cost.
...y more for more atoms than bits. Because atoms obviously and intuitively have a marginal cost.
...for advertisers… but it does make sense if they can do it with practically zero marginal cost. Atoms have marginal costs that are many orders of magnitude larger than bits.
...to n-1 generation models. The n generation frontier models will be available at marginal costs. That allows customers to self-discriminate into the more expensive models. As the models become cheaper to serve, they open up to more customers.
...the visual assets. LLMs make it orders of magnitude cheaper, but it still has a marginal cost. One way to do it is to have it feel open-ended and custom to the user… but subtly direct users to the same paths. Then, you can generate one hifi, d...
...nvert what had to be opex before into capex. Pay a fixed cost once and now that marginal cost evaporates. Tools that you use always have this shape. Writing didn't necessarily have this use as a tool, because the person you want to be affected...
...nds on this distribution. At least, for subscriptions to things with meaningful marginal cost. If every subscriber consumed 100% of their allowed use, the system would fall apart.
...e fixed cost of finding and doing the logistics for a project now dominates the marginal cost of executing the project. All rind, no pulp!
LLM marginal cost will stay high even as token cost declines. That's because of Jevon's Paradox. As the cost gets cheaper, we'll use them for even more things. Anythin...
...od of the community" is hard to motivate more than a few idealists on. AI has a marginal cost, which means there will already need to be infrastructure, perhaps we could use that.
I agree with this analysis that fixed-fee AI subscriptions can't work. The marginal cost is too high, and the value created by them is too high for users to ration their usage.
...l phone networks. Extremely capital intensive to build out, but then much lower marginal cost to operate. Though inference has much more marginal cost than cell phone networks. The actual service depends somewhat on quality, but they're all wi...
...t need to raise $1B Who's in?" ARR being the most important metric assumes zero marginal cost. That was a good assumption in the world of Saas. But it's a bad assumption in AI!
...he physical storage space to hold it, you have continued access to it with zero marginal cost. Superficially looks the same, but one is way more fragile than the other!
...ding tools that charge a high margin for hosting of the final app (despite zero marginal cost), but not for the creation of the app (despite significant marginal cost).
One of the reasons tech took over the world was zero marginal costs. But that doesn't apply to LLM-powered tools!
... necessary but not sufficient condition for user-alignment. Is it just the zero marginal cost that leads to engagement maxing? Attention is all you need… a neverending force of gravity not just for training LLMs but for business models too.
When a product has a non-trivial marginal cost but is free, then there's some ulterior motive that might arbitrarily misalign with your interest.