The most prestigious law school admissions discussion board in the world.
|
 |
 
|
AI wrapper "companies" going to ZERO when token prices triple
Poast new message in this thread
Date: February 19th, 2026 10:40 PM Author: Fluffy honey-headed main people lol no way is open ai and anthropic going to keep the free ride going forever (http://www.autoadmit.com/thread.php?thread_id=5836559&forum_id=2\u0026mark_id=5310893#49681739) |
 |
Date: February 20th, 2026 12:52 PM Author: brindle stirring shrine old irish cottage 🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨 |
Date: February 19th, 2026 10:43 PM Author: Aquamarine domesticated bawdyhouse They’re already going to zero man the wrappers are so fuckin done here (http://www.autoadmit.com/thread.php?thread_id=5836559&forum_id=2\u0026mark_id=5310893#49681746) |
|
Date: February 19th, 2026 10:45 PM Author: Fluffy honey-headed main people yeah good point. but also other kinds of AI startups are going to ZERO when token prices go up (http://www.autoadmit.com/thread.php?thread_id=5836559&forum_id=2\u0026mark_id=5310893#49681749) |
|
Date: February 19th, 2026 10:47 PM Author: Aquamarine domesticated bawdyhouse When do you think they’re gonna do the rug pull and 10x token prices (http://www.autoadmit.com/thread.php?thread_id=5836559&forum_id=2\u0026mark_id=5310893#49681754) |
|
Date: February 19th, 2026 11:11 PM Author: Ocher bat shit crazy crackhouse gross margin on tokens is already pretty good (around 50%), and it's a competitive market requiring billions in R&D and training to keep your tokens worth anything (http://www.autoadmit.com/thread.php?thread_id=5836559&forum_id=2\u0026mark_id=5310893#49681777) |
|
Date: February 19th, 2026 11:15 PM Author: doobsian adventurous alpha Models are only going to get cheaper to run relative to their strength, there’s far too much competition. The shit coming out of China on a relative shoestring budget (and price to run) is very impressive. Anthropic is rushing to build products on top and establish a dominant position in B2B early on for that exact reason
AI wrappers go to zero not because of token prices but because they only get easier to build the better the models get, and the frontier labs are obviously always a step ahead in access to the best models and talent |
|
Date: February 19th, 2026 11:19 PM Author: Aquamarine domesticated bawdyhouse How tf do you figure that models are only getting cheaper to run relative to strength when all of the gains in capabilities in the past 18 months have come from throwing as much inference compute as possible at whatever you want the model to do (http://www.autoadmit.com/thread.php?thread_id=5836559&forum_id=2\u0026mark_id=5310893#49681785) |
|
Date: February 19th, 2026 11:25 PM Author: Fluffy honey-headed main people maybe they are assuming that if raw comopute per query inccreases, efficency per outcome is still increasing faster overall? (http://www.autoadmit.com/thread.php?thread_id=5836559&forum_id=2\u0026mark_id=5310893#49681795) |
|
Date: February 19th, 2026 11:35 PM Author: Aquamarine domesticated bawdyhouse I think ultimately we’re going to end up at a spot with LLMs where you can keep getting marginal capability gains by spending tons of inference compute but they’re increasingly inefficient gains relative to cost
I worry that like capital itself, access to the most powerful AI capabilities will end up gated behind prohibitively high costs and access to “borrowed” Monopoly money. The rich get richer, the poor get poorer, etc |
|
Date: February 20th, 2026 1:00 AM Author: Aquamarine domesticated bawdyhouse https://x.com/gdb/status/2024662197692223857 (http://www.autoadmit.com/thread.php?thread_id=5836559&forum_id=2\u0026mark_id=5310893#49681868) |
|
Date: February 20th, 2026 5:14 AM Author: doobsian adventurous alpha Extreme compute scarcity is the only scenario in which our GPU capex frenzy stands a chance in the long run relative to China and its insurmountable lead in meatspace supply chain, of course they are going to say this. (http://www.autoadmit.com/thread.php?thread_id=5836559&forum_id=2\u0026mark_id=5310893#49681995) |
|
Date: February 20th, 2026 5:09 AM Author: doobsian adventurous alpha "all of the gains in capabilities in the past 18 months have come from throwing as much inference compute as possible at whatever you want the model to do "
Because that's just obviously not true? Minimax 2.5 is far more capable than Sonnet 3.5 or whatever |
|
Date: February 20th, 2026 12:52 PM Author: Aquamarine domesticated bawdyhouse i mean it's an exaggeration, yes. not literally all of the gains have come from throwing more inference compute at the prompt, but most of them have, for sure (http://www.autoadmit.com/thread.php?thread_id=5836559&forum_id=2\u0026mark_id=5310893#49682914) |
|
Date: February 19th, 2026 11:21 PM Author: Fluffy honey-headed main people the trend is toward cheaper tokens yes. but people would still pay for the frontier if they tripled prices so I wouldn't rule it out yet. plus its not profitable right now open ai and anthropic are basiclaly giving people a free ride. I think the main thing stopping them is that open source models are improving and distillation makes "good enough" very cheap. but still i think if they keep the gap wide enough at the frontier the prices could rise especially in industries that are high stakes (http://www.autoadmit.com/thread.php?thread_id=5836559&forum_id=2\u0026mark_id=5310893#49681790) |
|
Date: February 20th, 2026 1:53 AM Author: 180 ultramarine nursing home organic girlfriend This is why I'm furiously building a local llm with amd 9 and 4090 this weekend; then preordering M4 ultra with maximum RAM config on day 1 (http://www.autoadmit.com/thread.php?thread_id=5836559&forum_id=2\u0026mark_id=5310893#49681900) |
|
Date: February 20th, 2026 12:53 PM Author: brindle stirring shrine old irish cottage 🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨
🚨 this is a WLMAS account 🚨 |