Live data from Hacker News

Mistral reports on the environmental impact of LLMs

mistral.ai

11–20 of 59 posts

Re: Mistral reports on the environmental impact of LLMs

#11
post #9

Earlier quoted context omitted.

The difference is that food is important and live-giving and LLMs are a very fancy magic 8-ball

You don't need beef, beef is a lifestyle choice. I use LLMs to do all of my coding these days, it's certainly more essential for feeding me than beef.

Also a lifestyle choice

Re: Mistral reports on the environmental impact of LLMs

#12
post #11
post #9

Earlier quoted context omitted.

You don't need beef, beef is a lifestyle choice. I use LLMs to do all of my coding these days, it's certainly more essential for feeding me than beef.

Also a lifestyle choice

Yes but one with a much much much smaller impact as we just demonstrated.

This is exactly the kind of cognitive dissonance in people that I meant.

You literally see the math and go "but I like my meat, why should I give that up if you got your AI".

Because, as I just demonstrated, my AI takes a infinitesimal fraction of your meat.

It literally takes you only going vegan for a day to offset your entire AI usage of a year.

Re: Mistral reports on the environmental impact of LLMs

#13
post #3

I would really like it if an LLM tool would show me the power consumption and environmental impact of each request I’ve submitted.

And each toilet flush you make should also have a Co2 calculation which should go against your daily carbon allowance.

This is a good point because being curious about energy usage is the same thing as advocating for an imaginary rule about energy usage

Re: Mistral reports on the environmental impact of LLMs

#14
post #3

I would really like it if an LLM tool would show me the power consumption and environmental impact of each request I’ve submitted.

And each toilet flush you make should also have a Co2 calculation which should go against your daily carbon allowance.

Toilets are already labeled with their usage rate.

Re: Mistral reports on the environmental impact of LLMs

#15
post #12
post #11

Earlier quoted context omitted.

Also a lifestyle choice

Yes but one with a much much much smaller impact as we just demonstrated. This is exactly the kind of cognitive dissonance in people that I meant. You literally see the math and go "but I like my meat, why should I give that up if you got your AI". Because, as I just demonstrated, my AI takes a infinitesimal fraction of your meat. It literally takes you only going vegan for a day to offset your entire AI usage of a y…

This is spot on because there can’t be two issues that exist simultaneously. There can only be one thing that wastes enormous amounts of energy and that thing is beef

Re: Mistral reports on the environmental impact of LLMs

#16

I would really like it if an LLM tool would show me the power consumption and environmental impact of each request I’ve submitted.

That would be ... thousands of time less useful than giving you the same information at the motor fuel pump. Unfortunately this isn't one of those situations where every little bit counts. There are 2 or 3 things you can do to reduce your environmental impact and not using chatbots isn't one of the things.

Re: Mistral reports on the environmental impact of LLMs

#17
These conclusions are broadly compatible with "The Carbon Footprint of Machine Learning Training Will Plateau, Then Shrink" or, as I prefer, the PDF metadata title that they left in there, "Revamped Happy CO2e Paper".

Despite the incredible focus by the press on this topic, Mistral's lifecycle emissions in 18 months were less than the typical annual emissions of a single A320neo in commercial service.

https://arxiv.org/pdf/2204.05149

Re: Mistral reports on the environmental impact of LLMs

#18
post #6

So, using the smallest model for the task would help, as expected. A very small model could run on device to automatically switch and choose the right model based on the request. It would help navigate the difficult naming of each model of each vendor for sure.

> ” A very small model could run on device to automatically switch and choose the right model based on the request.”

This is harder than it looks. A “router” model often has to be quite large to maintain routing accuracy, especially if you’re trying to understand regular user requests.

Small on-device models gating more powerful models most likely just leads to mis-routes.

Re: Mistral reports on the environmental impact of LLMs

#19
What is the levelised cost per token? As in how we calculate levelised cost of energy.

If we take the total training footprint and divide that by the number of tokens the model is expected to produce over its lifetime, how does that compare to the marginal operational footprint?

My napkin math says per token water and material footprints are up 6-600% and 4-400% higher respectively for tokens on the order of 40B to 400M.

I don't have a good baseline on how many tokens Mistral Large 2 will infer over the course of its lifetime, however. Any ideas?

Re: Mistral reports on the environmental impact of LLMs

#20
post #12

Earlier quoted context omitted.

Yes but one with a much much much smaller impact as we just demonstrated. This is exactly the kind of cognitive dissonance in people that I meant. You literally see the math and go "but I like my meat, why should I give that up if you got your AI". Because, as I just demonstrated, my AI takes a infinitesimal fraction of your meat. It literally takes you only going vegan for a day to offset your entire AI usage of a y…

This is spot on because there can’t be two issues that exist simultaneously. There can only be one thing that wastes enormous amounts of energy and that thing is beef

You can try to misconstrue and ridicule the argument, but that won't change the math that if you have one thing that causes 1 unit of damage, and another thing that causes 100.000 units of damage, then for all intents and purposes the thing that produces 1 unit of damage is irrelevant.

And any discussion that tries to frame them as somewhat equally important issues is dishonest and either malicious or delusional.

My guess, as I've expressed earlier in the comment chain, is that it's emotionally easier for people to bike-shed about the 0.01% of their environmental impact, than to actually tackle things that make up 20%.

And no it's not only beef (which is a stand-in for meat and diary), another low hanging fruit is also transport, like switching your car for a bike.

But switching from meat and diary to a vegan diet would reduce up to 20% of your personal environmental impact, in terms of CO2.

And about 80-90% of rainforest deforestation is driven directly or indirectly by livestock production.

So it's simply the easiest most impactful thing everyone can do. (Switching your car for a bike isn't possible for people in rural areas for example.)

Post reply on HN