Sabitlenmiş Tweet
Jason
97 posts
Jason
@jaso1024_
Math & CS @Yale | kernels & optimization
Katılım Mayıs 2025
19 Takip Edilen19 Takipçiler
@ptr_to_joel better than most humans given enough time, try forcing it to make a mini-compiler, easier for it to reason through
English
reddit as a main source is crazy + this last source actually supports my argument (gives estimates between 0.003–0.007 gallons)
There are other (valid) reasons to hate AI, like it making people dumber, but water is one thats been relatively well researched, and most people informed on the topic tend to believe that its not that big of a deal
English
@jaso1024_ @duckonaroll @PureRinFunction If you want more, your figure about how much water uses AI, is indeed for a single prompt, not the water footprint of making, training and using AI, here:
ie.edu/insights/artic…
English
Why would this be a fair comparison. When comparing these sorts of things on a per-unit basis the generally accepted measures are average total cost (in which burgers use more) and average variable cost (in which burgers use more). The measure ur suggesting is to cherrypick some variables from the burger side and some from the AI side, which any intro econ prof would lambast you for
English
@jaso1024_ @Tokumostsanefan @PureRinFunction No, you are counting all the water consumed to make a burger, not to fry it, frying a burger would be the comparison you would need to make with prompting an AI once
If you want to keep the 600 Gallons, count the whole water usage to make an AI
English
@Re51221264 @PureRinFunction yeah, but by the thousandth query I think most ppl would give up and use their brain
English
@jaso1024_ @PureRinFunction Yes, but chatgpt query will make you do more chatgpt queries whil will lead to even more queries because of llm hallucinations
English
water usage to build a data center (the thing you've been talking about, and is also negl relative to coolant water use) \neq Coolant water usage (the thing I've already accounted for). It looks like a problem at face value, but when you view it in context (relative to all the other things in the world that consume water) its not that large
English
@jaso1024_ @duckonaroll @PureRinFunction What? The 5 Millions Gallons per day is what a large Data Center use per day as coolant, which is indeed a pretty big problem
Read and understand the questions I'm making because it looks like you are deliberately not "understanding" what I'm saying
English
@Just_AnS_ @duckonaroll @PureRinFunction because they process many prompts? the per day figure comes from variable cost, which is the figure i used. The number you're giving doesn't even factor the water to build a data center (because this number is difficult to calculate)
English
@jaso1024_ @duckonaroll @PureRinFunction I said, if you want to say that producing a burger takes 600 Gallons, then make the same comparison, and calculate how much water it takes to build a Data Center
GenAI large data centers usually use 5 Millions gallons per day
lincolninst.edu/publications/l…
brookings.edu/articles/ai-da…
English
Implying an unfair comparison by using the term "single prompt" seems odd to me, its Cost of 1x burger vs cost of 1x prompt, including all variable resources taken to make the item(s) in question
accounting for how much water it would take to make a data center would be like accounting for the water it takes to grow the trees are used to build a fence for the cows, which seems a bit silly (and is also negl water usage relative to the recurring cost)
English
@Tokumostsanefan @jaso1024_ @PureRinFunction The number he is using is misleading, no one would believe frying a burger takes 600 Gallons of water, he is taking into account the water that the cow, production and vehicles use to produce it
And with AI it only accounts for a single prompt
It is a lame tactic
English
@Tokumostsanefan @PureRinFunction Its less a question of which is worse (to me) and more a question of, if you care this much then why don't you lambast people who consume a decent amount of mean
English
@Tokumostsanefan @PureRinFunction yes, but they almost certainly are not asking millions of questions, even if they made 10k queries it would still be 4x less water consumption than a burger
English
@cha5m_ @PureRinFunction bad take, u don't need burgers to survive lol (its also several ooms more than a gpt query), I would also argue that a gpt query is substantially more useful than a burger
English
@jaso1024_ @PureRinFunction You’re comparing literal food that humans need to survive with an LLM prompt and you don’t see how evil that is??
English
This value does account for the coolant, if one didn't account for it the water usage would be pretty much 0?
Then also, I was quite charitable as well, the water number I used comes from the UC riverside study, and they account for numerous indirect sources of water consumption
English
@jaso1024_ @duckonaroll @PureRinFunction You are counting the water to make the whole production and transport for a single burger but with AI you are only counting the prompt
You are deliberately misleading or you are dumb
Try to account for the water usage to build the several Data Centers AI use and as coolant
English
@duckonaroll @PureRinFunction a burger is 600+ gallons and a gpt query is 0.0014 gal (note that this isn't a singular forward pass, but the average full query)
English
@jaso1024_ @PureRinFunction chatgpt queries are more numerous per consumer of them than burgers though
English
@Tokumostsanefan @PureRinFunction yeah, but a burger is 600+ gallons and a gpt query is 0.0014 gal (note that this isn't a singular forward pass, but the average full query)
English
@jaso1024_ @PureRinFunction kinda true, but remember a THousands if not millions people use chat gpt at once. so although technically it is kinda true if you compare 1 burger and 1 query. but 1 person doesnt just use 1 query in chatgpt /session dont they?
English
Anthropic is usually very good at handling PR for product issues. But this one is not it.
If it was a cost-saving move to massively increase token price and increase dispersion btwn api vs non-api plans. Company should own that instead of trying to gaslight the thousands of users that feel the change viscerally from similar usage. Asking people to just downgrade to inferior products is pedantic and adds insult to injury.
Lydia Hallie ✨@lydiahallie
Digging into reports, most of the fastest burn came down to a few token-heavy patterns. Some tips: • Sonnet 4.6 is the better default on Pro. Opus burns roughly twice as fast. Switch at session start. • Lower the effort level or turn off extended thinking when you don't need deep reasoning. Switch at session start. • Start fresh instead of resuming large sessions that have been idle ~1h • Cap your context window, long sessions cost more CLAUDE_CODE_AUTO_COMPACT_WINDOW=200000 We're rolling out more efficiency improvements, make sure you're on the latest version. If a small session is still eating a huge chunk of your limit in a way that seems unreasonable, run /feedback and we'll investigate
English
If it looks like a duck, swims like a duck, and quacks like a duck, then it probably is a duck
Anthropic@AnthropicAI
New Anthropic research: Emotion concepts and their function in a large language model. All LLMs sometimes act like they have emotions. But why? We found internal representations of emotion concepts that can drive Claude’s behavior, sometimes in surprising ways.
English
@EntrepGrow Found most gpu arch resources to be long and boring, so I think this’ll be good for people interested in learning
English
@EntrepGrow Yeah, in essence, the goal is to build a gpu from scratch, starting from transistors, with increasing levels of abstraction (once u build it it’s “yours”)
English
wake up babe..anthropic did another mech interpretability research and this time about emotions in llms
Anthropic@AnthropicAI
New Anthropic research: Emotion concepts and their function in a large language model. All LLMs sometimes act like they have emotions. But why? We found internal representations of emotion concepts that can drive Claude’s behavior, sometimes in surprising ways.
English