After a year Generative EU, flagship productsMicrosoft, putting ads to products, trying to refund the cost by increasing prices and Data Center Rental Cancellation. Google makes similar actions by adding an inevitable AI features to service in the work area when prices increased.
Does the tide finally finally open a generative AI? The situation is not so simple. Technical companies are fully committed to new technology – but people are struggling to find ways to pay for it.
Last week, Microsoft was withdrawn to some of the planned Data center rental. Movit increased subscription prices by 45% of the company’s flagship 365 programs and left silently Version that supports the ad some of the products.
Tech giant CEO, Satya Nadella, recently proposed The AI ​​did not make much value so far.
Microsoft’s actions may seem singular in the current wave of the AI ​​turmoil, Openai’s leash announcements as $ 500 billion Stargate Information Center Project.
But if we look closely, nothing in Microsoft’s decisions shows the withdrawal of AI. On the contrary, we see a change in the strategy of AI’s profitable strategy by changing the costs of consumers in non-open ways.
Generative AI is expensive. Openai was alleged by the market leader 400 million active monthly usersIt burns money.
Last year, Openai brought $ 3.7 billion in revenue – but almost 9 billion US dollars were spent for net casualties around $ 5 billion.
Microsoft is Openai’s largest investor and is currently providing cloud computing services, so open expenditures also cost Microsoft.
What is the generative AI so expensive? Human labor is associated with an aside, two costs AI models: training (modeling) and results (using the model).
If the training is a prerequisite (often large), the costs resulting in the user base are growing. And how much the model is how great it is, how much to run away.
Small, cheaper alternatives
A survey in Openai’s most advanced models It is up to $ 1,000 Only in force alone. In January, Openai General Director Sam Altman also said that the company $ 200 per month is not profitable.
These signals do not only lose money using only free models, but also through subscription models.
Both training and result usually occur in data centers. The costs are high, because the chips needed to use them are expensive, but the wearing of electricity, cooling and hardware.
So far, AI progress has been achieved using everything. Open describes The latest update is like “a giant, expensive model”. However, there are many signs because it is no longer necessary to approach this scale.
Chinese company Depth When he revealed this year earlier this year, for a small part of the cost of training, he set up models that can be compared to Openai flagship products. In the same way, the AI ​​(AI2) and Stanford University claims to develop a model for the researchers of the Allen Institute for the University of Seattle Up to $ 50.
In short, the AI ​​systems, developed by technological giants, may not be profitable. The cost of building and processing data centers is a great reason.
What is Microsoft doing?
Generongs, Billions, Microsoft, who sank billions of AI, trying to find a business model that will make technology profitable.
During last year, Tech Hazui, Copilot Generative AI Chatbot integrated into products aimed at consumers and enterprises.
It is no longer possible to get any Microsoft 365 subscription without copilot. As a result there are subscribers To see significant price rides.
As we can see, operate generative AI models in data centers. Thus, Microsoft probably looks for ways to do more about the user’s own devices – where the user has a hardware and its payment for its work.
A powerful tip for this strategy A small button Microsoft has begun to put in their devices last year. In the valuable real estate of the QWERTY keyboard, Microsoft devoted the key to copylot keys and laptops on their computers who can process AI on the device.
Apple pursues a similar strategy. The iPhone manufacturer does not offer the majority of AI services in the cloud. Instead, only new devices offer AI capabilities, and are sold as a privacy feature that prevents your devices from traveling in another place.
To push the expenses away
In our pockets, in our tables or even our tables or even our wrists (so-called “words are the promise to carry out the work of generation EU in our smart hoursExternal calculation“Because the” edge of the network “occurs).
This can reduce generation EU’s carbon, heating and water trace, reduce energy, sources and waste of data centers. You can also reduce bandwidth requirements and can increase user confidentiality.
But there are lands. Edge Computing, despite the economic and environmental concerns, which prevents upgrades, requires consumer computing costs, driving requirements for new devices, new devices. This can be strengthened with a newer, larger generation AI models.
And there are more problems. Distributed e-waste recycling. In addition, if a device dictates how well the EU can be in your education parameters, there will be no playground level for users.
Although the edge computing seems more “centralized”, it can also cause hardware monopolies. If a handful of company manages this passage, decentralization may not be as open as it seems.
As the expenses of the AI ​​infrastructure increase and model development, the costs are becoming a strategy to apply for consumers to apply for AI companies. Although large enterprises such as government departments and universities can handle these costs, many small enterprises and individual consumers can fight.
Kevin WitzenbergerResearch worker, Genay Lab, University of Technology of Queensland and Michael RichardsonMedia associate professor, Unsw sydney
This article is republished Conversation Under the Creative Commons license. Read original article.
