ChatGPT loses money on every query their premium subscribers submit. They lose money when people use copilot, which they resell to Microsoft. And it’s not like they’re going to make it up on volume - heavy users are significantly more costly.
This isn’t unique to ChatGPT.
Yes, it has its uses; no, it cannot continue in the way it has so far. Is it worth more than $200/month to you? Microsoft is tearing up datacenter deals. I don’t know what the future is, but this ain’t it.
ETA I think that management gets the most benefit, by far, and that’s why there’s so much talk about it. I recently needed to lead a meeting and spent some time building the deck with a LLM; took me 20 min to do something otherwise would have taken over an hour. When that is your job alongside responding to emails, it’s easy to see the draw. Of course, many of these people are in Bullshit Jobs.
Entirely agree with that. Except to add that so is Dario Amodei.
I think it’s got potential, but the cost and the accuracy are two pieces that need to be addressed. DeepSeek is headed in the right direction, only because they didn’t have the insane dollars that Microsoft and Google throw at OpenAI and Anthropic respectively.
Even with massive efficiency gains, though, the hardware market is going to do well if we’re all running local models!
Alibaba’s QwQ 32B is already incredible, and runnable on 16GB GPUs! Honestly it’s a bigger deal than Deepseek R1, and many open models before that were too, they just didn’t get the finance media attention DS got. And they are releasing a new series this month.
Local, efficient ML is coming. That’s why Altman and everyone are lying through their teeth: scaling up infinitely is not the way forward. It never was.
That’s the business model these days. ChatGPT, and other AI companies are following the disrupt (or enshittification) business model.
Acquire capital/investors to bankroll your project.
Operate at a loss while undercutting your competition.
Once you are the only company left standing, hike prices and cut services.
Ridiculous profit.
When your customers can no longer deal with the shit service and high prices, take the money, fold the company, and leave the investors holding the bag.
Now you’ve got a shit-ton of your own capital, so start over at step 1, and just add an extra step where you transfer the risk/liability to new investors over time.
Theres more than just chatgpt and American data center/llm companies. Theres openAI, google and meta (american), mistral (French), alibaba and deepseek (china). Many more smaller companies that either make their own models or further finetune specialized models from the big ones. Its global competition, all of them occasionally releasing open weights models of different sizes for you to run your own on home consumer computer hardware. Dont like big models from American megacorps that were trained on stolen copyright infringed information? Use ones trained completely on open public domain information.
Your phone can run a 1-4b model, your laptop 4-8b, your desktop with a GPU 12-32b. No data is sent to servers when you self-host. This is also relevant for companies that data kept in house.
Like it or not machine learning models are here to stay. Two big points. One, you can self host open weights models trained on completely public domain knowledge or your own private datasets already. Two, It actually does provide useful functions to home users beyond being a chatbot. People have used machine learning models to make music, generate images/video, integrate home automation like lighting control with tool calling, see images for details including document scanning, boilerplate basic code logic, check for semantic mistakes that regular spell check wont pick up on. In business ‘agenic tool calling’ to integrate models as secretaries is popular. Nft and crypto are truly worthless in practice for anything but grifting with pump n dump and baseless speculative asset gambling. AI can at least make an attempt at a task you give it and either generally succeed or fail at it.
Models around 24-32b range in high quant are reasonably capable of basic information processing task and generally accurate domain knowledge. You can’t treat it like a fact source because theres always a small statistical chance of it being wrong but its OK starting point for researching like Wikipedia.
My local colleges are researching multimodal llms recognizing the subtle patterns in billions of cancer cell photos to possibly help doctors better screen patients. I would love a vision model trained on public domain botany pictures that helps recognize poisonous or invasive plants.
The problem is that theres too much energy being spent training them. It takes a lot of energy in compute power to cook a model and further refine it. Its important for researchers to find more efficent ways to make them. Deepseek did this, they found a way to cook their models with way less energy and compute which is part of why that was exciting. Hopefully this energy can also come more from renewable instead of burning fuel.
Right, but most of their expenditures are not in the queries themselves but in model training. I think capital for training will dry up in coming years but people will keep running queries on the existing models, with more and more emphasis on efficiency. I hate AI overall but it does have its uses.
No, that’s the thing. There’s still significant expenditure to simply respond to a query. It’s not like Facebook where it costs $1 million to build it and $0.10/month for every additional user. It’s $1billion to build and $1 per query. There’s no recouping the cost at scale like previous tech innovation. The more use it gets, the more it costs to run, in a straight line, not asymptotically.
No way is it $1 per query. Hell a lot of these models you can run on your own computer, with no cost apart from a few cents of electricity (plus datacenter upkeep)
ChatGPT loses money on every query their premium subscribers submit. They lose money when people use copilot, which they resell to Microsoft. And it’s not like they’re going to make it up on volume - heavy users are significantly more costly.
This isn’t unique to ChatGPT.
Yes, it has its uses; no, it cannot continue in the way it has so far. Is it worth more than $200/month to you? Microsoft is tearing up datacenter deals. I don’t know what the future is, but this ain’t it.
ETA I think that management gets the most benefit, by far, and that’s why there’s so much talk about it. I recently needed to lead a meeting and spent some time building the deck with a LLM; took me 20 min to do something otherwise would have taken over an hour. When that is your job alongside responding to emails, it’s easy to see the draw. Of course, many of these people are in Bullshit Jobs.
are you telling me i can spam these shitty services to lose them money?
OpenAI is massively inefficient, and Atlman is a straight up con artist.
The future is more power efficient, smaller models hopefully running on your own device, especially if stuff like bitnet pans out.
Entirely agree with that. Except to add that so is Dario Amodei.
I think it’s got potential, but the cost and the accuracy are two pieces that need to be addressed. DeepSeek is headed in the right direction, only because they didn’t have the insane dollars that Microsoft and Google throw at OpenAI and Anthropic respectively.
Even with massive efficiency gains, though, the hardware market is going to do well if we’re all running local models!
Alibaba’s QwQ 32B is already incredible, and runnable on 16GB GPUs! Honestly it’s a bigger deal than Deepseek R1, and many open models before that were too, they just didn’t get the finance media attention DS got. And they are releasing a new series this month.
Microsoft just released a 2B bitnet model, today! And that’s their paltry underfunded research division, not the one training “usable” models: https://huggingface.co/microsoft/bitnet-b1.58-2B-4T
Local, efficient ML is coming. That’s why Altman and everyone are lying through their teeth: scaling up infinitely is not the way forward. It never was.
That’s the business model these days. ChatGPT, and other AI companies are following the disrupt (or enshittification) business model.
Now you’ve got a shit-ton of your own capital, so start over at step 1, and just add an extra step where you transfer the risk/liability to new investors over time.
Theres more than just chatgpt and American data center/llm companies. Theres openAI, google and meta (american), mistral (French), alibaba and deepseek (china). Many more smaller companies that either make their own models or further finetune specialized models from the big ones. Its global competition, all of them occasionally releasing open weights models of different sizes for you to run your own on home consumer computer hardware. Dont like big models from American megacorps that were trained on stolen copyright infringed information? Use ones trained completely on open public domain information.
Your phone can run a 1-4b model, your laptop 4-8b, your desktop with a GPU 12-32b. No data is sent to servers when you self-host. This is also relevant for companies that data kept in house.
Like it or not machine learning models are here to stay. Two big points. One, you can self host open weights models trained on completely public domain knowledge or your own private datasets already. Two, It actually does provide useful functions to home users beyond being a chatbot. People have used machine learning models to make music, generate images/video, integrate home automation like lighting control with tool calling, see images for details including document scanning, boilerplate basic code logic, check for semantic mistakes that regular spell check wont pick up on. In business ‘agenic tool calling’ to integrate models as secretaries is popular. Nft and crypto are truly worthless in practice for anything but grifting with pump n dump and baseless speculative asset gambling. AI can at least make an attempt at a task you give it and either generally succeed or fail at it.
Models around 24-32b range in high quant are reasonably capable of basic information processing task and generally accurate domain knowledge. You can’t treat it like a fact source because theres always a small statistical chance of it being wrong but its OK starting point for researching like Wikipedia.
My local colleges are researching multimodal llms recognizing the subtle patterns in billions of cancer cell photos to possibly help doctors better screen patients. I would love a vision model trained on public domain botany pictures that helps recognize poisonous or invasive plants.
The problem is that theres too much energy being spent training them. It takes a lot of energy in compute power to cook a model and further refine it. Its important for researchers to find more efficent ways to make them. Deepseek did this, they found a way to cook their models with way less energy and compute which is part of why that was exciting. Hopefully this energy can also come more from renewable instead of burning fuel.
Right, but most of their expenditures are not in the queries themselves but in model training. I think capital for training will dry up in coming years but people will keep running queries on the existing models, with more and more emphasis on efficiency. I hate AI overall but it does have its uses.
No, that’s the thing. There’s still significant expenditure to simply respond to a query. It’s not like Facebook where it costs $1 million to build it and $0.10/month for every additional user. It’s $1billion to build and $1 per query. There’s no recouping the cost at scale like previous tech innovation. The more use it gets, the more it costs to run, in a straight line, not asymptotically.
No way is it $1 per query. Hell a lot of these models you can run on your own computer, with no cost apart from a few cents of electricity (plus datacenter upkeep)