Decrypting OpenAI's $540M Expenditure on ChatGPT: Where Every Penny Goes

0:00Today on the podcast we are going to be talking about the cost of training chat GPT, how much OpenAI is spending on their entire company, how much they historically have spent and what is happening in the future. We're going to go over some of the costs, financial, time, and one you probably didn't think about, water. And we're also going to talk about what other companies are doing to help mitigate these costs, who some of the people spending the most money on, and why Microsoft and OpenAI are going to start making some new financial models and some new products in the very near future to help offset some of these costs.

0:35So without further ado, let's jump into the podcast. I think the biggest thing we want to talk about today is the fact that there's a big headline out of the information that says OpenAI's losses doubled to$540 million as it developed ChatGPT. So it's pretty crazy. As they were building out ChatGPT, apparently they had to hire a bunch of key employees from Google. And these employees can be very, very expensive. Some of them are being paid up to$5 million a year. and when you're a company as big as open ai and your product is absolutely pivotal and essential and based on these ai experts you're just going to pay whatever it costs especially when you've already received you know a billion dollars from microsoft and other investors and you're just going to spend whatever it takes to be able to get your product to market so apparently they were spending over 540 million dollars last year to get chat gpt ready developed and launched um and what's interesting is this is you know a previously unreported number um but a bunch of people from within the company i guess leaked that um and it's really a lot of this goes into just the steep costs of training these machine learning models um and before you know chat gpt launched this was a huge cost as a lot of this stuff had to be done from scratch and it was a massive undertaking so even after they've launched and the revenues kind of picked up reaching an annual pace of hundreds of millions of dollars um which is what they've done just a few weeks after they launched their paid version of chat gpt in february the costs of running their entire company i don't really think are going to slow down.

2:23I think they're going to keep rising as they're getting more customers and as they continue to train more and more advanced artificial intelligence models. And this technology is obviously rapidly developing and they are the ones that are paying for it and having to stay really on top of it. So their CEO, Sam Altman, he privately suggested that OpenAI AI may try to raise as much as$100 billion in the coming years. That is absolutely insane. So far, their biggest checks that have been written to them, Elon wrote some pretty big multi-million dollar checks at the very beginning. Some say of$100 million.

3:02Microsoft gave them a billion dollars. And then most recently, Microsoft gave them another$10 billion dollars which brings Microsoft up to about a 50 percent owner of the company and so the fact that they're looking at raising 10x up to 100 billion dollars in the coming years I think really just speaks to the fact that they think it's going to cost a lot more to continue to develop these capabilities as you can see they need to spend a lot of money on image they have dolly too but it's getting you know pretty soundly beaten by mid-journey at the moment one can only assume that they're going to be getting more into audio and more into video in the future and so as they kind of start looking at everything they want to do this is going to get incredibly expensive so very interesting they might be looking to spend or looking to raise a hundred billion dollars i think this is pretty important because right now we have a lot of other people creating these kind of ai models we have google working on bard we have bloomberg and recently Bloomberg said that they used 1.3 million hours of GPU time to help build their Bloomberg GPT large language model and so all of these are incredibly expensive at the moment and everyone's kind of right now there's almost like this big AI fight or arms race for GPUs server power from NVIDIA and from other platforms because everyone is trying to train these big huge models because it's such a hot topic right now.

4:29And so as they're fighting over it, the costs are going up even more. I think in addition, another cost that we really have to factor into all of this on top of just the training and the talent that ChatGPT historically and OpenAI historically spent is the fact that people trying to train new models are gonna have to start spending more on the data. ChatGPT kind of got under the radar because they were gonna be an open source platform. They had the keys to the kingdom when it came to Twitter and APIs to a ton of different companies. Reddit included that just allowed them to use the data for free and now that everyone is starting to monetize these platforms companies like Stack Overflow and Reddit have both said that if you would like to use their data you have to now pay them.

5:10So ChetGPT may have gotten in at a time when they were getting this kind of content on the cheap and it's left to be said if companies are going to retroactively make OpenAI pay for some of the data that they previously trained on. one other area that not a lot of people think about when they think about the costs of training these models is the water required so a recent um a recent round of research just came out saying that training chat gpt required enough water to fill a nuclear reactor's cooling tower um i think it's really important to know that uh in this new research they said that gpt-3 alone so that's before GPT-3.5 and GPT-4 consumed about 185 ,000 gallons or 700 ,000 liters of water.

6:00And so that's an average of every single conversation that someone is having with Chad GPT is equivalent of them dumping out a large bottle of fresh water onto the ground. So the people in the study obviously are emphasizing the environmental concerns with this, talking about the historic droughts that the U.S. is facing right now, and the fact that companies like Google training Google Bard right now have all of their servers and cloud facilities in states like Texas that are, you know, even more prone to drought, and it actually requires more water because there's more evaporation. I think in this study, they specifically focused not on water that was recycled because these servers, right, they recycle water through them to help cool them down, but specifically water that was you know evaporated or lost um and while i do think it is important to look at especially when we're looking at droughts and you know cycling or using water from local rivers and streams i do think that um this is probably less of an issue than i don't know nuclear waste as this article title alludes as they bring in nuclear reactors you know to try make it seem really radioactive and scary i mean this is water it evaporates um i don't know what else to say about it.

7:15I think we're all going to survive, although it is an interesting thing to keep an eye on as, you know, more and more AI models come out and they all need more water to make sure that, you know, we're using that and scaling that in a responsible manner. Now, all this to say, there are a ton of costs, right? We have the costs of the employees, the costs of training, the costs of now your data and the cost of your water, I guess. And finally, the final thing that I wanted to talk about is the fact that OpenAI and Microsoft are now implementing new ways to recoup some of these costs they're spending, new product lines.

7:51And one of the biggest ones that is happening right now is that Microsoft may begin to offer private chat GPT to businesses. As you know, a number of different businesses have banned their employees from using chat GPT. Most recently, Samsung had a bunch of employees who leaked confidential company information and source code to ChatGPT. And when I say leaked, I'm assuming they were developers who are saying they're pasting code in and saying, hey, based off this code, what can I do, blah, blah, blah. Because I have developers from my companies that do that. And I know a lot of different people that are getting development advice and code advice from ChatGPT.

8:26And so this happened to Samsung. sudden they banned all of their employees from using these ai models and they said they're going to work on some internal tools which i don't know how good they will be but that's what their plan is so because of that um companies microsoft specifically and in microsoft's because they gave they invested 10 billion dollars into opening i they got a special deal where they're allowed to resell um opening eyes products and so you see that with with um bing and you see that with a lot of their other products are incorporating it but specifically they're literally allowed to resell chat gpt to companies privately so private would be a company would be able to go and buy a special version of chat gpt on exclusive servers that are completely cut off none of the data that goes into it will ever be used for anything it's all on the company's specific account and it won't ever be incorporated into any future training data now the fact that they're doing this also makes you think that what you're currently telling chat gpt if you're not on this plan is being used for the training data but that's another conversation chat gpt and opening i also have hinted that they will be launching a similar product to this in the near future so i think between microsoft and open ai this is going to be a new way that they are trying to recoup some of their costs and making money back from this as companies like verizon chase city group and goldman sachs all took similar steps at banning their companies from really using this uh these ai models for fear of leaking private or sensitive information and so i think that once these companies that have banned these pretty big companies are able to let their employees use it there's going to be a massive boom in the consumption of these products it's going to make a lot of money for open ai and it's also going to add a lot of innovation and i believe a lot of automation for these companies that are allowing their employees to now start using it this is going to be a really interesting area to watch if you liked the podcast today make sure to subscribe to our newsletter there's a link in the description make sure to download the self-pause ai life coach app absolutely amazing for helping when you get whatever you're trying to get done accomplished, focusing on your goals, and also make sure to join us on our Facebook group.

10:29We talk all the time about what is happening in AI. It's a really awesome conversation that a lot of people are having. A lot of amazing insights are shared. We'd love to see you there. Thanks so much for listening to the podcast. And as always, have a great rest of your day.