@simon nope.

@simon I'm not sure if it makes sense to reason along the lines of compensation only. I mean, you would have to take into consideration public value created through your use of genAI, ressources (time, energy) saved by using genAI instead of x google searches and hours spent on stackoverflow. This won't be any means make genAI eco-friendly but I've got the impression that this is a more useful approach.

@simon I ran the numbers for excluding mammals from my diet, and it tends to offset CO2 from my personal flights. Some very rough calculations put a cheeseburger on par with about 12 hours of GPT use. So maybe. See https://suffolklitlab.org/protective-randomness-artificial-intelligence/ (the back of the envelope calculations are way down at the bottom, best to ctr-f "CO2")

@simon well, you could commit you and your offspring to store safely some of the nuclear waste coming out of Google's projected reactors. If it contains Plutonium, you're in for 50000 years.

@simon I think that the way forward here isn't to try and write it off as personal responsibility. Instead, put the focus where it belongs -- on the tech companies building, training & supplying models.

Demand (or better, regulate) that any power used for AI / LLM must come fully from investment by said companies into renewable energy sources, and that they must build out 25% extra capacity to feed back into the grid, beyond what they use.

@simon Cost is a pretty direct an fair/accurate comparison of CO2 emissions for this (and everything really). A flight is mostly the cost of burning fuel, and LLM usage is mainly the cost of electricity. You're almost certainly not using more than a plane ticket's worth of electricty in a year of LLM usage. Datacenters are also more likely to get their electricity from low carbon sources, which works even more in their favor.

@simon wait what, you have air trips every year? Let alone MULTIPLE air trips? Ok, sure, financially supporting LLM training and using cloud LLMs is probably not a significant addition to *your* resource consumption... 😕

@simon People seem to be especially concerned about water usage, but I feel like that can be straightforwardly converted to an electricity/CO₂ estimate using the equivalent number of BTUs provided by heat pumps based on the local energy mix. It ignores some trends (the overall mix is clearly trending towards more renewables over time), but likely good enough for a thought experiment.

@simon If you are looking at an overall CO2e budget for the way you live, then in principle you could look at where to allocate that 'spend' of CO2e

In economy a flight is approx 0.13kg CO2e per km flown. Double that for Premium economy and triple it for business class. LHR JFK return in economy is 1.6T CO2e per person.

There's estimates on tinterweb that suggest GPT CO2e at about 24tCO2e per day. Training cost est at 522tCO2e

TL;DR travel is always the big CO2e emission for most of us

@simon It depends. Electricity =! carbon, the source of power matters. If the source of power is from renewables, then the carbon cost of AI is much lower than what most algorithms account for.

@simon One could try to time the query when the renewables are abundant on grid, if one knew where the service is hosted.

@simon

https://piktochart.com/blog/carbon-footprint-of-chatgpt/

According to this, a return flight from SF to Seattle is equivalent to 92593 queries.

I don’t actually vouch for their math, especially since “queries” are such a broad unit which don’t take into account the model, or token count, or the carbon efficiency of the power source. But I think it’s a useful order of magnitude starting point.

@simon no

@simon I've found the self-reported Llama2 training (!) emissions interesting. See section 2.2.1 here:

https://arxiv.org/abs/2307.09288

That doesn't say much about inference, tho, but it's a start.

You'd also have to include not just GPU energy but e.g. emissions of the data center and supporting software around it, cluster utilization, the emissions of people working on this, traveling, and so on.

There's Scope1/2/3 etc. categories to make this a bit more structured but in the end it's tough.

@simon The part where you calculate your use‘s footprint makes sense (and would be very interesting to me, too) - justifying it by offsetting it against a specific unrelated „bad footprint“ activity like flying not so much unless you set a hard CO2 budget for yourself.

@simon Knowing the location of datacenters and using a source like https://app.electricitymaps.com/map you can guestimate the carbon footprint. My employer did the same for its datacenters in Norway but its somewhat complicated by a complex energy exchange market (we're connected to UK, Germany, Netherlands, Denmark, Sweden).