DeepSeek's R1 model training expenses dash the grandeur of big tech's substantial AI investment
In a groundbreaking development, Chinese AI developer DeepSeek has made a significant impact in the tech world with its flagship model, DeepSeek R1. This reasoning model, designed to excel at complex tasks such as mathematics and coding, has been downloaded over 10 million times on AI community platform Hugging Face.
DeepSeek R1's training process is unique, compared to a child playing video games, learning through trial and error. It uses a carrot-and-stick approach to reinforcement learning, where correct problem-solving is rewarded, and incorrect answers are penalised. This approach has enabled the model to develop its own strategies without copying human tactics.
The training process of DeepSeek R1 was completed using 512 Nvidia H800 chips, renowned for their high efficiency in data and energy use. DeepSeek's claim of cost-efficient training processes has been vindicated with the peer-reviewed publication of R1 in a reputable journal like Nature.
Reports from Wired suggest that DeepSeek's CEO confirmed the cost of training the R1 model to be 'more than $294,000' during a 2024 MIT event. This figure is significantly lower compared to the projected costs for a new model by Anthropic CEO Dario Amodei, who estimated it to be upwards of $100 billion in mid-2024. OpenAI CEO Sam Altman previously hinted that foundation model training cost upwards of $100 million.
DeepSeek's decision to offer R1 as an open weight model, freely available for anyone to download, has been welcomed by industry stakeholders. This move is expected to drive further innovation and collaboration in the AI community.
The tech industry giant, Cisco, is looking to capitalise on the 'DeepSeek effect' by leveraging the cost-efficient and innovative approach of DeepSeek R1's training process. AI reasoning models like DeepSeek R1 are purposefully trained on real-world data to 'learn' how to solve specific problems, making them valuable assets in various industries.
Researchers at Carnegie Mellon University have compared DeepSeek R1's training process to a child playing video games, learning through trial and error. This comparison underscores the potential of DeepSeek R1 to revolutionise the way AI models are trained and could pave the way for more cost-effective and efficient AI solutions in the future.
Read also:
- User Data Analysis on Epic Games Store
- AI Inspection Company, Zeitview, Secures $60 Million Funding for Expansion
- Ongoing trade friction as the American administration levies fresh import taxes on goods arriving from China
- Tech Titan Google Surpasses $3 Trillion Market Cap as Gemini Ousts ChatGPT, Boosting Google's Chat Platform
