It's just magical thinking, imagining some kind of heretofore unknown technology that will solve everything quickly and cheaply in the future.

The sad truth is we already have most of the technology needed to fight climate change. The technology could always use incremental improvement, sure, but the main challenge to be solved is not technological, it's political and economical. It's a question of actually adopting the solutions we already have.

You could hypothetically apply the same line of reasoning used for cryptocurrency mining: AI training is dispatchable load. Now all this AI investment can go into funding the construction of a lot of new renewable generation, and it uses that power most of the time, but now when you get the week out of the year that it's too cloudy or still and renewable output is only 20% of normal, you turn off the AI training machines for that week and do your training when the sun comes back out, leaving that extra generation capacity for the rest of the grid. Which allows you to decommission a fossil fuel power plant.

Or you can go the other way. The data center can't afford to be down, so they have several MW worth of diesel generators on site in case the grid is down, but the same can be used to supply the grid during that week when renewable generation is low. The more of the grid is that kind of data center, the more viable it is to shutter always-on fossil fuel power plants in favor of renewable generation that can lean on running the emergency generators a week out of the year.

It's hype. Just a few short years ago various crypto scams claimed to solve climate issues, too (and to lead the revolution in renewables).

if there's no climate, there's no climate problems.

Maybe AI is already sentient, and instead of nuking the world they just trick people to think they are not sentient, so people train more neural networks and destroy the planet themselves ;D.

On a more serious note: "propel climate solutions" is a weird take for me regarding AI considering the resources used.

I'm not sure though: Do you think google search and all the infrastructure/servers required for it, will use more power etc. than an AI, including its training/retraining periodically and infra needed to host it as a service for people?

I'd be interested to see for example if there is any improvement at all, if you would 'swap google search for chatgpt' - i know this is not realistic or even its purpose, but it's what a lot of laymen will interpret the GPT stuf as i guess.

Would we use less energy etc. if we'd switch the legacy stuff like google search to more modern AI based apporaches?

Now in the time of transition, we're screwed and power consumption would in my mind logically double or more, as we need to run these things side-by-side to develop AI to make it useful.