You might be making money using it, although the historic data you cited seems pretty poor for that argument, but what if investors pulled out of OpenAI, novelty users stopped buying, and localities started charging them more for power and bandwidth using the excuse that subsidized goods shouldn’t be wasted on them? What if they had to raise their rates from $30 to $300 and you could save money by hiring a part timer at $29 for those 10 hours of work?
I think, but I don’t know for certain, that OpenAI is not economically feasible and are only offering services at a cost because more and more investment keeps pouring in, and because by quietly lowering level of quality to the majority of their users they can save pennies on the dollar as they have been shown to do.
I’ve tested it with chatgpt, meta.ai, and two locally run models. What I’m using it to do isn’t particularly complex so they’ve all worked, even if the big models stopped the ones that ran on my own hardware would still be saving me time. They just take a little longer to run, on the order of 10-20 seconds rather than the 2-5 I’m seeing online.
The big version of the open source models can be run on rented servers for cheaper than $300 per month, so I’m not worried about your hypothetical situation at all.
You might be making money using it, although the historic data you cited seems pretty poor for that argument, but what if investors pulled out of OpenAI, novelty users stopped buying, and localities started charging them more for power and bandwidth using the excuse that subsidized goods shouldn’t be wasted on them? What if they had to raise their rates from $30 to $300 and you could save money by hiring a part timer at $29 for those 10 hours of work?
I think, but I don’t know for certain, that OpenAI is not economically feasible and are only offering services at a cost because more and more investment keeps pouring in, and because by quietly lowering level of quality to the majority of their users they can save pennies on the dollar as they have been shown to do.
I’ve tested it with chatgpt, meta.ai, and two locally run models. What I’m using it to do isn’t particularly complex so they’ve all worked, even if the big models stopped the ones that ran on my own hardware would still be saving me time. They just take a little longer to run, on the order of 10-20 seconds rather than the 2-5 I’m seeing online.
The big version of the open source models can be run on rented servers for cheaper than $300 per month, so I’m not worried about your hypothetical situation at all.