Hacker News new | past | comments | ask | show | jobs | submit
> One of the things driving the cost is having to keep training new models. It was every 6 months or so, now it seems like every 2-3 months. If they stop the other companies will catch up, so they are forced to pace (or race) the frontier to make sure they always have something better to offer if/when another company catches up.

Exactly. Plus new models need more and more training data, so it will just keep costing more and more, there's no end to it (except when they run out of data, which is likely too and soon).

> This is likely why they are pushing into other areas like mathematics and more recently genetics -- they need new customers to keep and increase the token spending.

Yes, but there are not much new areas left to invade, they are running out of possibilities pretty quickly. There's only so much areas, and they have already covered most of it.

All in all, their costs are keep increasing while their chance to continue tokenmaxxing and grow income is drastically decreasing. Not a good sign.