Comment by davydm
4 days ago
the ai companies have a problem at the moment: the compute required for inference (especially) costs a metric shit-ton of money - so they're crowd-sourcing the funds (unsuccessfully so far, tbh - for example, a $20/m claude sub that i'm on for work, and I regularly see individual conversations where the tokens cost more than that - even up to $350.
It's a tool tho, not a panacea. With the right guidance, for example, you can get some good code results.
With the right guidance
Catch 22 --- In order to pay for itself, AI needs to eliminate the very jobs that build the experience and expertise necessary to provide the "right guidance".
At the same time, this will restrict production of quality content that is needed for model training.
In other words, AI creates a self degrading/defeating feed back loop.
Absolutely. At its core, AI is just another algorithm that is subject to the same 'Garbage In-Garbage Out' principle. As more and more of the data used to train the models, is itself generated by AI; the feedback loop you describe gets stronger.