Gary Marcus argues the AI industry faces airline-like economics with razor-thin margins and massive operating costs

@GaryMarcus The way out is by making better models, and making them faster. Read Jevons paradox. Your ideas always assume a static universe.

@GaryMarcus How did Netscape go from multi-billion dollar dot com glory to AOL glory-hole firesale? Their product was made available for free by another company..

I don’t think the airline analogy holds because changing airlines is pretty frictionless.

switching models, especially if you’ve started building harnesses, tools & scaffolding around one isnt as frictionless.

Experienced this ourselves when trying to switch from Claude to codex. So I’m not sure if it’s going to play out quite the same as the airlines.

@GaryMarcus The earth simply does not have the resources to power AI chips & robots forever. After a brief golden period, it will have to be rationed at exorbitant prices. It's only "efficient" in the short-term. Much easier to create humans, all you need is an egg, sperm, sunlight & food.

@GaryMarcus Noble to put life before money…Thats why they are winning and the US is losing. Wake up.

@GaryMarcus China also is trying to sell hardware like NVIDIA will get ready around 2030 so they destroy the US ai companies using Open source models cause you don't need the best model to do everything and sell you a cheap better hardware to use the model local, companies will prefer this.

@GaryMarcus So with recent USG intervention the total addressable market for frontier AI models has shrunk from a planet of 8bn, down to a handful of trusted organizations. Will that affect their revenue?

@GaryMarcus airline analogy is brutal but makes sense

trillions in compute and still no one really printing except nvidia

@GaryMarcus Like the Airlines - consumers benefit but not the companies that built the network and run the planes. Profitable but not crazy software margins.

@GaryMarcus Except that it's unclear if all the stranded assets will actually benefit infrastructure much. So not like the internet and railroad bubbles.

@GaryMarcus race to bottom

@GaryMarcus People would willingly surrender their critical thinking in exchange for comfort, ultimately embracing the very systems that limit their freedom.

@GaryMarcus China doesn't run windmills to power it, big difference.

@GaryMarcus How would anyone know when this is entirely new?

@GaryMarcus Money is in the application layer.

@GaryMarcus sumpah ini mah

@GaryMarcus FOMO and hype drove everyone I reckon.

To exces it appeared very powerful.

Those who should have known better were getting paid a lot.

@GaryMarcus That's a great analogy Gary, and that's not even considering the risk of a new paradigm away from the transformer architecture era

@GaryMarcus Xi Jingpin has all the money he needs, there are other priorities

@GaryMarcus Big picture - What's starving the AI industry? Data. Open models will win on data alone = better models = beating the US in the "AI war". Then they make the biggest, most capable models for Government use only.