You're right to be concerned about the climate impact of deploying these systems. I understand where you're coming from, and I have the same concerns too.
The tools that we use have a cost, just like heat in our offices. But that cost allows us to have a positive impact. It's up to us to weigh the tradeoff.
The cost is less than you'd think: A ChatGPT average query uses as much electricity as ten Google searches, or about 0.34 watt-hours. That’s about what an oven would use in a little over one second, or a high-efficiency lightbulb would use in a couple of minutes.
LLMs (training and data centers) have significantly added to the total electricity demand, but they can be more efficient than humans for some applications. Imagine answering a question with FixBot or hiring a mechanic to drive out to your house.
But local efficiency gains don’t diminish the broad impact. US data centers consume 4% of energy now, and are growing rapidly. There is a race to run these things off clean power sources, and we need to accelerate that as fast as we can. In the meantime, we have a challenge.
iFixit's incentive is to save cost and run these models as cheaply as we can. The cost of these models is 1:1 lined up with the environmental cost: the more we pay model providers, the more they spend on energy producing tokens for us.
We strive to walk a fine line in the middle, balancing environmental impact with positive world impact. Repair is good for the planet, and the more accessible we can make repair, the better. Over time, algorithmic improvements are rapidly driving down the cost and impact of the sort of applications we're doing now.
We’re going to monitor our impact closely.
With regards to enshittification: FixBot is instructed and designed to link references to our content wherever possible. One way to think about FixBot is that it is an expert at navigating and using iFixit. Our library is huge and it can be complex to identify your device and find the right information. We carefully made FixBot into the best and fastest user of iFixit, setting up a number of native "tools" for it to use iFixit to the fullest.
It is not a new Large Language Model (LLM). Training a new foundational model costs tens of millions, and would not be a good environmental idea even if we could afford it.
While the way that we manage these systems is new, the approaches for managing quality in a computer system are not new to us. We have a long track record of publishing quality software, using automated testing to rigorously validate each change before we publish it. We're applying that same level of rigor to FixBot.
I understand that there are a lot of reasons to be skeptical of AI. FixBot is a way for us to help people fix things that we wouldn't have been able to help before. That's our north star: teaching the world to repair. FixBot answered a question correctly in Farsi today. I don't think iFixit has ever helped people in Farsi before. That's pretty exciting.
1 
100 
9