Matisyahu's Blog

After having a nasty week of bad weather it is good to see the start of a new week which will mean geting back into a routine again – I’ve set my alarm and I know it’ll be difficult for the first couple of days but getting into a good routine will mean that everything else falls into place. I’ve set the alarm to 9:30am and in the morning (although I may push it back to 10:30am depending how I feel when I turn off the light and close my laptop) I’m going to head down to the supermarket to pick up some groceries with the $75 voucher from work plus the $15 voucher I got by accumulating points as I shopped.

I’ve been following the whole AI hype cycle and it appears that reality is setting in – the idea of ‘all you can eat’ subscriptions are not economically viable and as Ed Zitron noted on a recent podcast regarding the need to move to a system where you pay as you go. The problem is, at least as far as I see it, the move to a pay as you go model might make sense if the models were actually good enough to the point that you don’t waste tokens.

The problem is that there are examples of companies blowing through a whole years budget in a matter of a few months – the end result is something that doesn’t actually save money the moment you start paying as you use it and that is something that companies are quickly facing. I can’t help but get the feeling that many of the AI companies were hoping that their technology would become so critical that businesses would be willing to pay through the nose for it but the reality is that if the numbers don’t add up then aren’t going to keep paying for it.

I think what you’re going to find is that after the hype dies that down we’ll see that eventually it’ll be primarily aimed at businesses, AI companies will probably eventually jettison chat just as they jettisoned video creation on demand because it is too costly and has no chance of actually turning a profit given how much they would need to charge vs how much organisations are willing to pay. I wouldn’t be surprised if you end up having a developer tools with preconfigured set of commands that call an API to ensure that the least amount of tokens are used when when completing the command rather than relying on a wordy prompt.