Logan

Severe caffeine addict and enjoyer of fine culture

[Logan @logancsack]

@Kris_Plugg @VladTheInflator Lol who are your ancestors and what did they accomplish?

[Logan @logancsack]

@RupertLowe10

[Logan @logancsack]

@NoahRyanCo Other countries have found that it's easier to send their citizens to the US to work and send money back than it is to develop their own economies

[Logan @logancsack]

@AbbyJohnson Since 1990, at least 2.2x more people have died from abortion than people who died as a result of war since the dawn of time

[Logan @logancsack]

In 1824, Thomas Jefferson called coffee “the favorite beverage of the civilised world."

The Ottoman empire banned coffee in an attempt to stave off rebellion.

Theodore Roosevelt drank as much as a gallon of coffee a day.

Caffeine is the drug of choice for high agency individuals.

I just wish TJ was around to experience white monster.

[Logan @logancsack]

@ApoStructura https://t.co/0VTWsO53DO

[Logan @logancsack]

The AI water usage panic is a hoax.

There are plenty of valid criticisms of AI (thanks Pope Leo), but the water thing is blown out of proportion.

Most of these data centers use closed-loop cooling, meaning that once they fill the system with water, that water continues to be recycled indefinitely.

Very little new water gets added as the data center runs.

Even data centers that use evaporative cooling use a small amount of water compared to other industries.

The average ChatGPT query uses about 0.000085 gallons of water, roughly 0.00032 liters.

MMLU, one of the standard AI benchmarks, has about 15,908 questions.

So one full MMLU-style eval is roughly:

15,908 × 0.00032 L = about 5.1 liters of water.

Now compare that to clothing:

A cotton T-shirt requires about 2,700 liters of water throughout its lifetime to make it and wash it.

A pair of jeans: nearly 3,800 liters.

Shirt + jeans: about 6,500 liters.

That means one shirt + jeans is roughly equal to:

20 million average ChatGPT queries.

Or about 1,270 full MMLU benchmark evals.

A fast-fashion cotton outfit can carry more water footprint than hundreds or thousands of serious AI benchmark runs.

[Logan @logancsack]

@kayleighmcenany And it's been this way since 2010

[Logan @logancsack]

Napoleon ended the godless French Revolution, ended the French Republic, crowned himself emperor, and reestablished Catholicism as the official state religion.

While making himself an emperor, he also installed his brothers as monarchs of Spain, Holland, and Westphalia

[Pamphlets @PamphletsY]

@putxiwhipped5 Napoleon KILLED GOD

Absolute Monarchy was vanquished by Napoleonic Reforms, history on horseback spoke hegel and TRULY!

the CORRUPT PAPACY was sent asunder plunging into the reliquiary of history, power then came from below and reason and enlightenment kicked in europes belly

[Logan @logancsack]

The AI water usage panic is a hoax.

There are plenty of valid criticisms of AI (thanks Pope Leo), but the water thing is blown out of proportion.

Most of these data centers use closed-loop cooling, meaning that once they fill the system with water, that water continues to be recycled indefinitely.

Very little new water gets added as the data center runs.

Even data centers that use evaporative cooling use a small amount of water compared to other industries.

The average ChatGPT query uses about 0.000085 gallons of water, roughly 0.00032 liters.

MMLU, one of the standard AI benchmarks, has about 15,908 questions.

So one full MMLU-style eval is roughly:

15,908 × 0.00032 L = about 5.1 liters of water.

Now compare that to clothing:

A cotton T-shirt requires about 2,700 liters of water throughout its lifetime to make it and wash it.

A pair of jeans: nearly 3,800 liters.

Shirt + jeans: about 6,500 liters.

That means one shirt + jeans is roughly equal to:

20 million average ChatGPT queries.

Or about 1,270 full MMLU benchmark evals.

A fast-fashion cotton outfit can carry more water footprint than hundreds or thousands of serious AI benchmark runs.

[Remarks @remarks]

JUST IN: 🇺🇳 UN warns AI could soon use more water than everyone in the world needs to drink.

[Logan @logancsack]

@ProjectConstitu @MrsErikaKirk Imagine being a grown adult who doesn't capitalize the word "I"

[Logan @logancsack]

@RonDeSantis Patriot 🫡

[Logan @logancsack]

@felixhhaas Lovable

[Logan @logancsack]

@Lovable yessir 🫡

[Logan @logancsack]

I was just talking about self-directed coding via CLI and IDE. Which is what I think Jack was talking about in the original tweet.

But for OpenClaw, I'm not sure what to think about the future of using provider plans via OAuth. I think they (Anthropic for sure) want to keep their plans as more of a consumer subscription, and make sure that applications like OpenClaw are hitting the API.

I'm not sure that using a plan via OAuth is even the best way to utilize OpenClaw anyway.

Using Claude models for every OpenClaw action is wasting precious tokens on a model that is overkill.

Recently, I've been very intentional in making sure that I route requests to different models depending on what they're good at.

I'm actually building a layer to route actions to the appropriate model to get max cost-efficient performance.

For example, using GPT 5 nano in place of Haiku 4.5 for heartbeats can result in 93+% cost savings.

But I still want Opus for OpenClaw tasks that require max intelligence and high level planning.

Thus the importance of a smart router.

[Logan @logancsack]

I forced myself to majorly cut back my AI sub spend a few months ago, so I went to the $20/mo with each provider.

I'm only going to allow myself to have one $200/mo going forward, based on which model is my daily driver each month.

I think the important point is that I get more productive Codex 5.3 usage on $20 than Opus 4.6 on $200.

In my case, Opus is cost prohibitive to use for anything other than high-level planning

[Logan @logancsack]

@jackfriks @postbridge_ would be 🔥

[Logan @logancsack]

@Rasmic Based. Amen brother