Well big chunk of the electricity cost are fixed cost, meaning if enough people lower their bill with PV and batteries then prices will have to rise for their connection. In fact some suggest that it would make more sense to let people pay a flat rate.
Read Ross Anderson book Security Eng., it deals a lot with these types of issues. In short, banks try not do anything about as long they do not have to pay for it.
I suspect the issue is similar here with the multi stakeholder problem here. The college needs as many paying students as possible, the workload of the staff should be as low as possible, the office giving out loans wants to have a low workload... All in all good scammers will serve all these needs and happily take the money in the process.
If you think about marginal cost, such experiments can be run almost at only the cost of extra electricity used for that computation, which in Europe is often zero, at least by the ones who own the compute.
Ok, but now getting rid of your legacy Cobol system becomes 10x cheaper, hence replacing it becomes feasible. Deploying an industrial robot with sensors to do specialized complicated thing x is now only a few skilled queries away and costing you only one workday. Writing software to replace manual process which n many workers do and would lead to savings but not enough to justify a normal programmer writing the code...
if price of code goes down, short term fixes become worth the money.
This is a reminder to the hacker news community, that most likely if you increase the efficiency of anything the demand will go up, not down. So taking this to Developer-AI Cyborgs, being as productive as n-many Developer in 2022: the demand for these Cyborgs will be higher than for pre Cyborg developers.
No, this does not work. Solar thermal will not work without power. Any battery solution monitoring your home temperatures are enough. Then you can travel home or ask your neighbor.
But the LLM interacts with the program and the world through debugger, run-time feedback, linter, fuzzer etc., we can collect all the user feedback, user pattern ... Moreover, it can also get visual feedback. Reason through other programs like physic simulation etc. Use a robot to interact with the device running the code physically. Can use proof verifier like lean, to ensure its logical model of the program is sound. Do some back and forth between the logical model and the actual program through experiments. Maybe not now, but I don't see why the LLM needs to be kept in the Chinese Room.
I mean, probably many assume that we drive around our kids everyday (which we dont) cause they can walk or bike to school. Then that on every small trip they get a phone or tablet (neither true). It is just for 1 hour or more trips which are once a month or less.