It’s been an exciting week in the realm of AI with the debut of DeepSeek, an open-source LLM that toppled ChatGPT from its top spot. The battle for AI large language model supremacy continues and during the course of under a week, a new contender appeared and became the new champ, was accused of cheating, found to have a security vulnerability, and has gone on to be supported by multiple industry giants, including one whose stock prices fell due to the new AI’s debut.
Debuts and disrupts
The new AI from China unveiled its free open-source LLM in December. It was claimed that the lab needed only two months to develop the model and cost under $6 million. By comparison, it was said that ChatGPT led to roughly $540 million in losses for its developer OpenAI, back in 2023. Meanwhile, DeepSeek R1 was released this week and immediately turned the AI industry upside down. Tech stocks dropped following its release but NVIDIA took the biggest plunge in its history with nearly $600 billion in losses on the same day, setting a new record for the U.S. stock exchange.
Potentially using banned tech and rival model data
It didn’t take long for suspicions to arise regarding how the Chinese lab achieved its results. The lab has stated that it used NVIDIA H800 chips for its R&D, something that NVIDIA has since confirmed and further elaborated that roughly 2,000 chips were used. The NVIDIA H800 is a chip compliant with U.S. microchip export restrictions but there is an investigation into whether or not the lab may’ve used tech that is restricted from being exported to China. OpenAI has also stated that it has evidence of the lab using its model to train DeepSeek.
Not the most secure?
It’s debatable if any chatbot could be considered secure given the methods that they use for training but some might find it a bit of a shock to find out how readily available data is for users of the new AI. Wiz Research found an unsecured database housing “million lines of log streams containing chat history, secret keys, backend details, and other highly sensitive information.”
Wiz Research did notify the lab who in turn took action to secure the ClickHouse database. However, it remains unknown how much data may’ve been extracted before being locked down. The lab did indicate on Monday that it had also been the target of a Cyberattack.
From rival to supported service
NVIDIA has been quick to seek a turnaround following the release of DeepSeek R1. While most of the tech world witnessed the release of its GeForce RTX 50 series yesterday, the GPU manufacturer also announced support for the latest version with its NIM microservice.
Per NVIDIA:
“Developers can experience the DeepSeek-R1 NIM microservice, now available on build.nvidia.com.”