Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Google's carbon emissions surge nearly 50% due to AI energy demand

cnbc.com

Post reply on HN
Print view View original

111–120 of 134 posts

Re: Google's carbon emissions surge nearly 50% due to AI energy demand

#111

Post by muglug »

williamcotton wrote: post #108
muglug wrote: post #106

Earlier quoted context omitted.

It was a glib comment, but I fundamentally reject the premise that we'll have more personal time because of ChatGPT. It's like saying that buying a faster computer that compiles quicker means you can spend more time with your family. Theoretically true, but in practice people who are that way inclined can find myriad other ways to use the new technology to eat up time.

Let's unravel what you've said here. You're saying that there is no way that I'm not a bad parent because I would have found a way to ignore my family regardless of some time saving device. Glib? That's not glib. That's just being a jerk. And fueled by what, your personal distaste for LLMs? People who are that way inclined?

I don't think you're a bad parent, but I do think you were using hyperbole when describing how much time ChatGPT saved you. I also don't believe that you would spend a week away from your kids to make a 1,200-line app that just explains how simple bash commands work. https://explainshell.com/ already does a pretty great job of that.

I don't have a personal distate for LLMs — I use ChatGPT weekly and I work for a company that ships LLM-powered functionality that I use daily. But your comment implied that LLMs can today provide a big productivity improvement for developers, and that’s not something I think there’s any concrete evidence for.

I do find the hype immensely distateful, especially in response to an article pointing to the environmental downsides of that excitement.

Re: Google's carbon emissions surge nearly 50% due to AI energy demand

#112

Post by williamcotton »

muglug wrote: post #111
williamcotton wrote: post #108

Earlier quoted context omitted.

Let's unravel what you've said here. You're saying that there is no way that I'm not a bad parent because I would have found a way to ignore my family regardless of some time saving device. Glib? That's not glib. That's just being a jerk. And fueled by what, your personal distaste for LLMs? People who are that way inclined?

I don't think you're a bad parent, but I do think you were using hyperbole when describing how much time ChatGPT saved you. I also don't believe that you would spend a week away from your kids to make a 1,200-line app that just explains how simple bash commands work. https://explainshell.com/ already does a pretty great job of that. I don't have a personal distate for LLMs — I use ChatGPT weekly and I work for a comp…

What I made is not like explainshell.com at all. It is a GUI for writing command pipelines. Updates to the prompt are parsed into an AST and used to update the GUI. Updates in the GUI are used to build an AST to update the prompt.

I wrote it so I didn't have to keep writing things like this into the line editor:

  pg -d test_database -c 'SELECT * from test_table;' | tsvtocsv | ggplot 'ggplot(df, aes(as.Date(date), value)) +
    geom_col(fill = "red") +
    labs(x = "date") +
    theme_minimal()' | pngtohtml
Instead, I could just type in the app:

  pg | tsvtocsv | ggplot | pngtohtml
And then fill in the blanks with an instance of the Monaco code editor, looking like:

https://github.com/williamcotton/guish/assets/13163/a5214e93...

It's a small program but it's not a trivial task to get the UX working in a productive manner. I took plenty of the same wrong turns that I would have if I was writing it by hand but was able to iterate much more quickly.

It's more exploratory in nature, seeing how text input and GUI input can complement each other in a novel manner. Sure, I could do this by embedding a SQL query in an R script or something but there's just something that rubs me the wrong way about embedding one language in another instead of having them work side by side.

It's not "just explaining how simple bash commands work".

So thank you for calling me a shitty parent and devaluing my work. You seem awesome.

Re: Google's carbon emissions surge nearly 50% due to AI energy demand

#113

Post by muglug »

williamcotton wrote: post #112
muglug wrote: post #111

Earlier quoted context omitted.

I don't think you're a bad parent, but I do think you were using hyperbole when describing how much time ChatGPT saved you. I also don't believe that you would spend a week away from your kids to make a 1,200-line app that just explains how simple bash commands work. https://explainshell.com/ already does a pretty great job of that. I don't have a personal distate for LLMs — I use ChatGPT weekly and I work for a comp…

What I made is not like explainshell.com at all. It is a GUI for writing command pipelines. Updates to the prompt are parsed into an AST and used to update the GUI. Updates in the GUI are used to build an AST to update the prompt. I wrote it so I didn't have to keep writing things like this into the line editor: pg -d test_database -c 'SELECT * from test_table;' | tsvtocsv | ggplot 'ggplot(df, aes(as.Date(date), valu…

I’m sure you’re a great parent, but you’re not a convincing advocate for the time-saving power of LLMs.

Either way I was wrong — this project does look like a cool idea, and I’m sorry for dismissing it entirely.

Re: Google's carbon emissions surge nearly 50% due to AI energy demand

#114

Post by williamcotton »

muglug wrote: post #113
williamcotton wrote: post #112

Earlier quoted context omitted.

What I made is not like explainshell.com at all. It is a GUI for writing command pipelines. Updates to the prompt are parsed into an AST and used to update the GUI. Updates in the GUI are used to build an AST to update the prompt. I wrote it so I didn't have to keep writing things like this into the line editor: pg -d test_database -c 'SELECT * from test_table;' | tsvtocsv | ggplot 'ggplot(df, aes(as.Date(date), valu…

I’m sure you’re a great parent, but you’re not a convincing advocate for the time-saving power of LLMs. Either way I was wrong — this project does look like a cool idea, and I’m sorry for dismissing it entirely.

Apology accepted. And I'm sorry for getting heated!

Re: Google's carbon emissions surge nearly 50% due to AI energy demand

#115

Post by ETH_start »

piva00 wrote: post #51
ETH_start wrote: post #35

Earlier quoted context omitted.

There is reason to be cautious about every kind of endeavor where the potential for enrichment can cloud judgment. However, the possibilities opened up by technologies this radical in the magnitude of the improvements they make in specific fundamental societal functions are enormous, and for that reason, they are on the balance massively beneficial, and that includes chat bots.

> However, the possibilities opened up by technologies this radical in the magnitude of the improvements they make in specific fundamental societal functions are enormous, and for that reason, they are on the balance massively beneficial, and that includes chat bots. The improvements it can be dreamed of to the bottom line of companies by replacing expensive labour are massive, that's why it got US$ 1tn Capex downed…

Reducing the need to expend expensive labor to produce goods/services is massively beneficial. It is the basis of all economic development and wage growth.

Re: Google's carbon emissions surge nearly 50% due to AI energy demand

#116

Post by namaria »

sva_ wrote: post #103
lagrange77 wrote: post #78

Earlier quoted context omitted.

> Not sure if energy use is a sufficient criteria to dismiss some technology. This universe progresses by turning low entropy energy into high entropy energy, which the use of energy is doing. What is the connection between those two sentences?

I think that things like intelligence and productivity are directly related to an increase in entropy. In this universe, we have a budget of entropy, and when it is used up, we're done for. We have to make our mark before that is the case, or the program that runs our simulation will just terminate with no result. It is a very pseudoscientific believe of mine, I guess.

> we have a budget of entropy

In closed systems. The Universe doesn't seem to be a closed system.

Re: Google's carbon emissions surge nearly 50% due to AI energy demand

#117

Post by yencabulator »

williamcotton wrote: post #71
Frieren wrote: post #27

I see the current gold rush of AI the same way I saw crypto-currencies. Even if originally there were people that believed in the concept it became just a snake-oil sellers business. The parallelism is made even more relevant by its hungry use of electricity. There is a future for AI but it is not what we see companies developing right now. Chat-bots are more dystopian and problematic than useful. AI future (and pres…

I'm getting very tired of the claim that LLMs are not useful. I hand-wrote perhaps 10 lines of code in this entire project: https://github.com/williamcotton/guish I spent most of the weekend playing with my kids in the pool. Three years ago this would have taken more than a week of my time and I wouldn't have done much else.

Great example of LLM-produced code. I found the first bug after 20 seconds of reading. Enjoy!

Re: Google's carbon emissions surge nearly 50% due to AI energy demand

#119

Post by seadan83 »

ETH_start wrote: post #115
piva00 wrote: post #51

Earlier quoted context omitted.

> However, the possibilities opened up by technologies this radical in the magnitude of the improvements they make in specific fundamental societal functions are enormous, and for that reason, they are on the balance massively beneficial, and that includes chat bots. The improvements it can be dreamed of to the bottom line of companies by replacing expensive labour are massive, that's why it got US$ 1tn Capex downed…

Reducing the need to expend expensive labor to produce goods/services is massively beneficial. It is the basis of all economic development and wage growth.

This can also describe the great enshittification race of every good/product/service. The 'wage growth' part does not always follow. Middle-skilled laborers are still be pushed out of the economy. [1]

[1] https://www.urban.org/sites/default/files/publication/31566/... (data is on pages 13, 20, 21 & 23)

Re: Google's carbon emissions surge nearly 50% due to AI energy demand

#120

Post by seadan83 »

brink wrote: post #43
josefritzishere wrote: post #3

The galling part of this is that the demand for AI is artificial. Consumers are not demanding AI.

Exactly. I was the first to adopt nearly every new technology in my family growing up. With AI, I query claude like twice a week for things that a search engine never gives to me straight like - "How long do I cook a rack of ribs?", but it's not something I'm about to pay $20/mo for, which is the only way that AI is sustainable. I use it sparingly, and I only use it because it's free. There's just not much appeal to…

If you don't know how long to cook a rack of ribs for - how will you know if the AI gave you a correct answer?

> There's just not much appeal to something as unreliable as AI.

Even your trivial example I would question regarding the unreliability.

AI is probably only best for qualitative things, like touching up photos, identifying data patterns for humans to then investigate. Places where 1% inaccuracy leads to bad tasting rack of ribs, or outright incorrect answers - the fact LLMs are able to produce reasonable sounding answers is amazing - but not necessarily useful.

Post reply on HN