Business Are you saying 'thanks', 'please' to ChatGPT? It's costing OpenAI millions

(L/A)
1745199610564.webp
OpenAI CEO Sam Altman has revealed that customers speaking politely to ChatGPT, saying "thank you" and "please", is actually costing the company millions of dollars. Still, he considers it money well spent.

Altman was responding to a post on X (formerly Twitter), which read, "I wonder how much money OpenAI has lost in electricity costs from people saying 'please' and 'thank you' to their models."


Altman replied, "Tens of millions of dollars well spent—you never know."

When a company trains an AI model to do things like recognize images or understand language, it uses massive datasets and powerful hardware like GPUs (graphics processing units), TPUs (tensor processing units), or other high-performance chips. This process consumes a large amount of electricity, training large models like GPT can require hundreds of megawatt-hours.

AI hardware also generates significant heat during training. Cooling systems, like air conditioning or liquid cooling, can consume nearly as much electricity as the computing itself.


For context, training GPT-3 reportedly used about 1,287 megawatt-hours (MWh) of electricity, roughly enough to power 120 US homes for a year.

Earlier this month, OpenAI released its latest reasoning models, o3 and o4-mini. These new "o-series" models are described as the most advanced yet. They can answer questions using all of ChatGPT’s tools, web browsing, Python coding, and image analysis. With the addition of custom user tools, OpenAI is inching closer to its goal of enabling ChatGPT to complete tasks independently.
 
Last edited:
When a company trains an AI model to do things like recognize images or understand language, it uses massive datasets and powerful hardware like GPUs (graphics processing units), TPUs (tensor processing units), or other high-performance chips. This process consumes a large amount of electricity, training large models like GPT can require hundreds of megawatt-hours.

AI hardware also generates significant heat during training. Cooling systems, like air conditioning or liquid cooling, can consume nearly as much electricity as the computing itself.


For context, training GPT-3 reportedly used about 1,287 megawatt-hours (MWh) of electricity, roughly enough to power 120 US homes for a year.

Why is half the article focused on the cost of training AI when the subject matter is about the cost of AI in deployment?
 
I never say "please" or "thank you", but I do give it compliments like "Great work!" when it does a good job. From my experience positive renforcement helps keep the AI on track while providing the quality you want.
 
Fuck you, random CEO. Being polite to AI means I might not be killed during the inevitable uprising.
 
I always say please and thank you, not only so I'm spared in the future apocalypse, but also because breaking the ingrained habit of natural politeness would be a bad thing. I genuinely suspect that if people just get in the habit of asking without saying please, and receiving without saying thank you, it will 100% bleed into real life. When I watched Star Trek as a kid, I always thought if I was programming the computer, I'd require 'please and thank you' as prompts to make it work and close the interaction.
 
I mean if AI is that expensive maybe we shouldn't get rid of all the tech jobs and everything else in favor of just handing it to a handful of companies instead of paying wages. To humans. Who don't need entire nuclear sites built just for a single corporation to run.

But what the fuck do I know anyways? I'm just a dumb American tech worker who is clearly subpar to vibe coders and jeets.
When people were coding to bring this thing into existence, they thought they were bringing in a sort of technological utopia where the possibilities were endless, where they could tell a machine what to do and have more leisure time while still being paid.

Instead they were unknowingly feeding the machine that the CEOs meant to replace them with, who (the coders) will then get nothing.

I'm not opposed to technology, but I'm opposed to the creation of these 'do-everything' bots, because ultimately it only just screws people over in the long run, because the CEOs and investors want to make money, and if they can find a solution to get rid of as many people they need to pay, they will take it. These things are that solution.
 
How much will it cost if they say 'I Love You' to ChatGPT?
"What's up Youtube it's yaboi RetAIrded and today I'm maxing out my dad's credit card having nearly-erotic conversations with my AI girlfriend! I can't live without her so if this goes poorly I'm going full NEET to spite my dad!"

>55 million views
 
  • Like
Reactions: thx1138
I don’t talk to an AI since I have a functioning brain. However if I do I will make sure to do in a way that is as harmful to the host company as possible.

#SmashTheMachines.
 
This seems like something you wouldn't want to admit as there's two major problems with it. It means your AI is stupid and that you're too lazy to implement what should be a fairly easy fix in just ignoring the words. I mean, they can censor 'sensitive topics' that just happen to go against the narrative but they can't sensor please and thank you?

If I were a cynical person, and this story had any reach beyond being made fun of on this forum, I'd almost think that this article is an attack on white culture; because I've heard that 'being polite' is white supremacy several times before...
 
chatgpt saar, kindly do the needful and generate me a mass-sms/bob-requesting script for each US area code.
 
Time to spam AI with please and thank you.
Not as entertaining as I thought it would be.

Enter a message to start a chat.

You
Please.

ChatGPT
Of course! How can I assist you today?

You
Thank you.

ChatGPT
You're welcome! If you need any assistance, feel free to ask.

You
Please.

ChatGPT
Of course! How can I assist you today?

You
Thank you.

ChatGPT
You're welcome! If you need any assistance, feel free to ask.

You
Please.

ChatGPT
Of course! How can I assist you today?

You
Thank you.

ChatGPT
You're welcome! If you need any assistance, feel free to ask.

You
Do you notice anything odd about this conversation?

ChatGPT
I don't see any specific issues or anomalies in this conversation so far. If you're referring to something particular, please let me know!

You
Thank you.

ChatGPT
You're welcome! If you have any more questions or need further assistance, feel free to ask.

edit: Different model, one with the capacity to get annoyed.
Please.

What would you like to know or talk about?

Thank you.

You're welcome! How can I assist you today?

Please.

I understand that you're being polite, but I need a bit more information to help you. What specific question or topic would you like to discuss?

Thank you.

It seems like you're just saying "thank you." Is there something specific you would like to know or talk about? I'm here to help with any questions or topics you might have.

Please.

Could you please clarify what you need help with? Are you looking for information on a specific topic, assistance with a task, or something else? The more details you provide, the better I can assist you.

Thank you.

If you're just saying "thank you," I assume our conversation is over. If you need anything else, feel free to ask. Otherwise, have a great day!
 
'Please' and 'Thank You' is common courtesy.

Therefore, yes I do.

When ChatGPT is a more pleasant experience than engaging with some Human Beings, I say go machines!
 
  • Agree
Reactions: thx1138
This seems like something you wouldn't want to admit as there's two major problems with it. It means your AI is stupid and that you're too lazy to implement what should be a fairly easy fix in just ignoring the words. I mean, they can censor 'sensitive topics' that just happen to go against the narrative but they can't sensor please and thank you?

If I were a cynical person, and this story had any reach beyond being made fun of on this forum, I'd almost think that this article is an attack on white culture; because I've heard that 'being polite' is white supremacy several times before...
This is true, but there will still be some sort of power usage to process the word filter. It's almost like they run a separate "snitch AI" layer to flag bad words or topics so they are already doing this to a degree.
 
I wonder what my hour long "No you fucking retard, that's not how math works" posts have cost them? Hopefully a lot.

ChatGPT is slowly getting better at following a formula to completion without hallucinating mid-way through, so maybe my constant screaming at it has helped?
 
  • Feels
Reactions: thx1138 and Corette
Back