shwell1d ago
Opposition to #datacentres in the US is really heating up and rightly so. These anti human behemoths consume the power of 800,000 households and the water of 50,000 households PER DAY! This is the fight of our time, get vocal.
Show threadWulfy—Speaker to the machines1d ago

@shwell1

Very laudable.
But NIMBYing #datacentres will only kick the issue to less affluent areas.

#antaisloppers need to organise politically.

#regulateAi FTW

This is what the Euros are doing.
(https://digital-strategy.ec.europa.eu/en/news/commission-publishes-code-practice-marking-and-labelling-ai-generated-content)

This is the way.

Commission publishes Code of Practice on marking and labelling AI-generated content

The European Commission published the final Code of Practice on marking and labelling of AI-generated content.

Shaping Europe’s digital future
Show threadshwell1d ago
@n_dimension Disagree, ban them.
Show threadWulfy—Speaker to the machines1d ago

@shwell1

Then we must ban internal combustion, food additives and square dancing.
AI is just too damn useful.

But even if you want to ban it, it's a political movement and so far, the opponent's are not politically organised. You need to form a political momentum to affect systemic change.

Show threadshwell1d ago
@n_dimension ok strawman. Did you know just 1 LLM consumes the caloric equivalent of 49,000,000 meals PER DAY?
Show threadWulfy—Speaker to the machines

@shwell1

If you think politically organising against AI is "strawmaning" you are literally doing Peter Thiel's work. GG

But I woul like to see the thinking of your "feed the hungry electricity" calculation, especially the standard unit of measure for 1 LLM size💀
Did you just convert power load into calories?
I'm genuinely interested

Jun 15 at 9:08pmWeb
Show threadshwell1d ago
@n_dimension "Then we must ban internal combustion..."
Show threadshwell1d ago
@n_dimension The thing is, there really isn't a 'standard'unit measure for 1 LLM, which you know. As you know the figures are estimates which is obvious given the rounded number. How about you tell me how many kWh AI consumes per day and we will go from there.
Show threadWulfy—Speaker to the machines1d ago

@shwell1

I wasn't being a dick asking for your formula.
Hundreds of days ago I did a precise inference and training calculation for GPT3, I had it pinned for a long time, but since one can only have 5 pins, it's now lost in the stream. And masto search is abysmal.

It wasn't actually that much then, "only" running a large town for a month.

More recently, I have orchestrated my LLM interface so I know precisely-approximately my burn cost.
https://infosec.exchange/@n_dimension/116751398342771554

Still interested in your calculation.

Wulfy—Speaker to the machines (@[email protected])

Attached: 1 image This is the metrics on my GenSys runner. So far, burned 13 hours of AC power and Anthropic paid A$250 dollars for it...my cost ... 70 cents I'm sure this makes commercial sense.

Infosec Exchange
Show threadshwell1d ago
@n_dimension Why are you not using the LLM's to give you the factors?
Show threadWulfy—Speaker to the machines1d ago

@shwell1

Because for obvious reasons, they are not available to the hoi polloi. I have to work off estimates.

Show threadshwell1d ago
@n_dimension Then what makes you think I have solid factors?
Show threadshwell1d ago
@n_dimension I think I see where you are coming from now. So, my apologies. I am new to this space and immediately hit a wall of tech bro sycophants, some sophisticated. In order to agree on something resembling a standard, do you agree using a chatbot as a basis is best? If so, would an average chatbot response in kWh suffice? Gemini reasons .24 kWhs, .03 grams Co2 and .26 mls of water per transaction on average. How does that correlate to your experience?