AI critic Gary Marcus called the current business model unsustainable
boom. there it is. “the costs of the tokens are far higher than actual value that these tokens are generated”.
unless that changes, this can’t possibly last.
"The costs of token are far higher than the actual value that these tokens are generating at scale," @Cisco's @jpatel41 tells @ReedAlbergotti.
"The big risk in the market is, if you don't create an equilibrium there, then people just pull back on using tokens."
Positive users call Google's AI token strategy genius and praise the executive's watch, while negative users see token utility as a recurring bottleneck and mock the business model as upside down.
We need to talk about @jpatel41's watch
"The costs of token are far higher than the actual value that these tokens are generating at scale," @Cisco's @jpatel41 tells @ReedAlbergotti.
"The big risk in the market is, if you don't create an equilibrium there, then people just pull back on using tokens."
@GaryMarcus That makes the OpenAI price cut even more of an issue.
One of three things happen:
Option 1:
- buy worse tokens much cheaper until OpenAI go bust - become reliant on Anthropic - Anthropic become token monopolists
This is likely why google is focusing more on cost than being the frontier model leader. I bet that google could have catalyzed the LLM craze before OpenAI but they simply didn't see how that would translate to higher profits. Search and Ads are/were cash cows because the per-request cost was ridiculously cheap. OpenAI forced them to commercialize their AI.
@MatthewBerman @jpatel41 Banger of a watch
@GaryMarcus What happens when we have models we can run locally that are 90% as good as the ones at the data centers? Honestly, if I owned a company, I'd just invest in local inference, and just for those employees that need it. Why send your IP to a data center and pay their rates?
@GaryMarcus 🤔
@spacemonkey4242 @GaryMarcus I think this was actually genius by Google (in hindsight).
> Watch OAI go red hot and then drop an anvil on itself > Learn what OAI and Ant did wrong/right and then > Beat the survivor with their own playbook OR hook onto the winner (they own 14% of Ant) > Cost optimise
@GaryMarcus lol it’s an upside down business model just like teslie
The cost of building your widget is more than the value of the widget or what the widget can generate in revenue
But hey, it’s gonna make magic and cure cancer and replace all jobs
@GaryMarcus To some extent that’s a skill issue. Does he somehow speak for … everybody getting inference tokens?
@GaryMarcus 2:
- buy zillions of crappy OpenAI tokens at giveaway price, and grant OpenAI a good-news story and a stay of execution - Anthropic drops their prices to meet OpenAI - price war that fundamentally knackers OAI, Ant or both; bubble pops - Google/Amazon/MS make out like bandits
@GaryMarcus 3:
- 1, but when Ant become token monopolists, everyone decides exactly how much the tokens are worth to them and how much AI they need - Switch to open weights/Gemini/Copilot because those come free, or with your phone/computer - Ant beat OAI - but fight the whole sector
@GaryMarcus token utility is still the bottleneck nobody wants to address
happens every cycle, then everyone acts surprised