The launch of DeepSeek’s AI chatbot has disrupted the tech industry, surpassing OpenAI’s ChatGPT as the most-downloaded app in the US. The app’s efficient “large language model” reportedly has reasoning capabilities similar to existing models but requires significantly lower costs to operate. This shift raises questions about the future of AI development, ultimately showcasing that sophisticated technology can emerge from smaller companies. As sustainability concerns play a role, the AI landscape may undergo further transformation.
DeepSeek, a fledgling Chinese company, has stirred up the tech scene with its new AI-powered chatbot, sending shockwaves through the stock market. The app quickly climbed the charts, snagging the title of most-downloaded free iOS app in the U.S. and contributing to a staggering $600 billion drop in Nvidia’s market value in just one day—a first in U.S. stock history. But what makes DeepSeek’s offering stand out?
The answer lies in the app’s advanced “large language model” (LLM), which reportedly matches the reasoning capabilities of established models like OpenAI’s, yet at a fraction of the cost. Dr. Andrew Duncan, director of AI at the Alan Turing Institute, notes that DeepSeek claims to have slashed the costs associated with training their model, R1, through various technical strategies that reduce computation time and memory usage. The base model V3 required around 2.788 million hours of training, costing under $6 million— a far cry from the $100 million-plus price tag mentioned by OpenAI’s CEO, Sam Altman, for training GPT-4.
Amid Nvidia’s market plunge, it’s worth pointing out that DeepSeek utilized approximately 2,000 modified Nvidia H800 GPUs for training. These GPUs adhere to U.S. export rules, suggesting DeepSeek possibly stockpiled them prior to tightened restrictions in October 2023. The company has seemingly had to innovate to maximize the resources available to them, working within these limitations.
The implications of this innovation extend beyond savings; they might help curb AI’s environmental impact. Data centers, which are typically power-hungry and require substantial water resources, contribute significantly to carbon emissions. A recent estimate suggests ChatGPT emits upwards of 260 tonnes of CO2 per month—comparable to running 260 transatlantic flights. Thus, making AI models more efficient is definitely a step in the right direction for sustainability.
Yet, it’s still uncertain whether DeepSeek’s models genuinely bring about energy efficiencies or if increased accessibility will just ramp up overall energy consumption. No doubt, the upcoming Paris AI Action Summit will face pressing questions about sustainable AI as our dependency on these tools grows.
What really draws attention is DeepSeek’s rapid ascent. Founded only in 2023 by Liang Wenfeng, who’s already being dubbed an “AI hero” in China, the company is already producing a competitive LLM. The open release of their model’s “weights” and accompanying technical documentation allows other developers to adapt and utilize it, fostering more collaboration and innovation.
This openness sets DeepSeek apart from other models like OpenAI’s, which are more or less black boxes in terms of accessibility. Researchers are now hard at work trying to gather the missing information about the data sets and code used for training.
Interestingly, some of DeepSeek’s techniques for reducing costs aren’t entirely new. They share similarities with methods employed in other LLMs, such as Mistral AI’s Mixtral 8x7B model. DeepSeek also uses the “mixture of experts” approach, letting specific tasks be handled by specialized sub-models. They’ve even been transparent about their unsuccessful attempts to enhance LLM reasoning through various strategies, filling researchers in on where to go next in refining AI capabilities.
Looking ahead, DeepSeek might be paving the way for a future where creating powerful AI models doesn’t require such hefty resources. As many experts have pointed out, if the costs of developing AI drop, both businesses and government could start adopting these technologies more readily, feeding demand for products and the chips that drive them, thus perpetuating the cycle.
In short, we might soon see smaller players like DeepSeek gaining ground in the AI sector, creating valuable tools that could significantly simplify our lives. Dismissing them would be a misstep.BEACON
For more insights and updates on technology, you can subscribe to our Tech Decoded newsletter, while our Essential List provides a curated selection of features delivered to your inbox every week. Follow us for more science, technology, and health stories across our social media avenues.
In summary, DeepSeek’s rise is intriguing, highlighting how smaller companies can shake up a landscape traditionally dominated by big players. The AI chatbot employs advanced techniques to reduce costs and may even address environmental concerns while fostering a more collaborative spirit in AI development. As they continue to navigate these waters, their impact on the industry warrants close attention, marking a potential shift towards more sustainable and accessible AI tools.
Original Source: www.bbc.com
