Cutting the Cost of Cutting-Edge AI
Reducing costs and optimizing flow in artificial intelligence: Strategies to lower expenses in AI technology
The world of AI is rapidly advancing, with low-cost generative AI technology based on large language models becoming increasingly accessible to developers. Major AI startups like OpenAI, Baidu, and DeepSeek claim to have slashed development costs for large generative systems by up to 100 times, using "secondary" and imprecise online data.
However, it's challenging to verify the veracity of these claims, as all information about the drastic drop in Large Language Model (LLM) development costs stems from forum statements and social media posts. One thing we know for certain is that services like ChatGPT remain unprofitable and require substantial financial investment, with investments in AI technologies soaring from $16 billion a decade ago to an estimated $320 billion annually now.
Key Insights:- AI startups are reporting substantial reductions in the cost of large generative systems.- The claims about cost reduction originated from online discussions.- Overall investments in AI technologies have increased dramatically in the past decade.
Despite the hype surrounding generative AI, non-generative solutions are proving more practical and useful for many real-world applications, such as smart video surveillance systems. These systems draw information from "living nature" - real-world business processes in manufacturing and urban management.
In Russia, for example, smart video surveillance systems are on the rise, with clients seeking to significantly increase their coverage. However, a common issue arises when the number of cameras grows exponentially, leading to a "techno-atavism" - where clients compare new intelligent video monitoring systems to traditional video recording services, fearing a rapid increase in costs.
Key Insights:- Non-generative AI solutions are proving more practical and useful.- Smart video surveillance systems are on the rise in Russia.- Clients are concerned about the escalating costs of increasing camera coverage.
To make AI affordable and widespread, a systematic approach is needed. Taking smart video surveillance systems as an example, we can implement a "breadth-first search," moving quickly in all possible directions to reduce costs. Here are some strategies:
Optimizing Hardware
Choose specialized chips designed for efficient processing, like GPUs or TPUs, or even switching to more cost-effective options like Sophgo's tensor processing units (TPUs). Remember, neural networks don't "watch" videos; they analyze still images, so the chip you select matters.
Reducing Video Storage Costs
Compress video streams to lower storage costs, and consider shortening the storage period for video recording, for example, from 30 to 5 days. Remember, the goal is to detect violations, not to store video footage for unnecessary periods.
Leveraging Neural Networks
Interconnect neural networks, so they "inform" and "support" each other, reducing the number of TPUs needed. Optimize neural network designs to capture specific events, such as infractions, and send those frames to the neural network.
Key Insights:- Implementing a "breadth-first search" can drastically reduce costs for AI systems, like smart video surveillance.- Choose specialized chips for efficient processing.- Interconnect neural networks to reduce the number of TPUs needed.- Optimize neural network designs to capture specific events.
By implementing these strategies, it's possible to create a complex yet scalable AI system that is uninterrupted, resilient, and cost-effective. This system can analyze violations with relative ease, while minimizing data collection and complying with privacy regulations like the 152-FZ "On Personal Data." Brilliantly simple, yet highly effective.
So there you have it! With some careful planning and strategic optimization, we can tap into the exciting world of AI without breaking the bank. It's time to embrace the future, one cost-effective AI system at a time.
- Investments in artificial-intelligence (AI) technologies have expanded significantly, moving from $16 billion a decade ago to an estimated $320 billion annually now, demonstrating the growing financial interest in AI.
- In the realm of business and technology, optimizing hardware with specialized chips, reducing video storage costs, and leveraging neural networks can lead to cost-effective AI systems, such as smart video surveillance, without compromising their practicality and utility.
