HN Explorer

Ask HN: How did the industry settle on weekly limits?

27 points by saratogacx ↗ HN
I understand that the cheap compute ride wouldn't last forever but something that feels somewhat unique seems to have come about from all of this AI belt tightening.

Weekly Limits

This policy seemed to come from almost nowhere but was quickly adopted across many products. Cutting off access for 25% of the time based product you've paid for feels like it is just incompatible with a subscription at the core conceptual level. Cooldown times and over-use back-offs are nothing new but there is a drastic difference in tone from 5 hours to 5-7 days.

I'm at a loss to see how this became an acceptable practice with the most common answer being "Buy more subs"

7 comments

[ 2.8 ms ] story [ 30.4 ms ] thread
bigyabai
You can still buy tokens direct from the API. The subscription model is not much more than a glorified paid trial.
tkiolp4
Just cancel your subscription.
dbeyzade
"Weekly caps feel like a psychological nudge toward higher tiers rather than a genuine resource constraint. The industry collectively normalized it because it works."
tughvn
It's probably a capacity problem, but we need to speak up for something better.
cyanydeez
there isnt something better. theyre money losing divices and unless they addict the user with promises of efficiency, they wont make money.

the market is distorted. more AI datacenters will be disasters for local ecosystems, but might serve a few more people.

however, if these things truly do the work of 100x they only need 1 guy . do you think youll be that guy?

almost everything in the gridt economy says if you cant throw money at it, youll be priced out.

if you dont have a local strategy, youll be SOL

dormento
I think one reason might be that the "horizon" of developments related to optimization and new techniques is very short. Since we're still having recurrent breakthroughs, new things are being attempted every other week. Too much money and investor attention means competition is beyond fierce, and the turn of innovation follows. I think this plays a part into the weirdly timed cooldowns, terms of service, inconsistent quality etc.

It also mimics the goldrush for training content. Acquiring content (just like gauging LLM capabilities) is too fast for even lawyers to keep up. Society as a whole kinda threw their hands up and hoped things would just solve themselves instead of the world suing itself into oblivion. So the threshold of acceptability has moved accordingly.

I don't like it but it is what it is :/