Podcast Episode
Multiple job postings released this month reveal plans for search functionality capable of processing text, images, and audio inputs. The company is simultaneously scaling its AI agent capabilities, with job listings detailing requirements for training data, evaluation systems, and dedicated platforms to support agents that operate continuously with minimal human intervention.
DeepSeek recently published a research paper describing a new training method called Manifold-Constrained Hyper-Connections, which analysts described as a breakthrough for scaling large language models. The company has provided limited information about its next-generation flagship model, though a recent reference to model1 appeared on its public GitHub account.
According to the letter, documents show Nvidia engineers helped DeepSeek optimise algorithms, software, and hardware, allowing the startup to train its models using roughly 2.8 million GPU hours on Nvidia's H800 chips, considerably less than what American developers typically require for comparable systems.
Nvidia pushed back against the allegations, stating it would be illogical for the Chinese military to rely on American technology. China's embassy in Washington criticised the allegations as politicising trade and technology issues.
DeepSeek Makes Bold Move into AI Search to Challenge Google and OpenAI
January 29, 2026
Audio archived. Episodes older than 60 days are removed to save server storage. Story details remain below.
Chinese AI startup DeepSeek is hiring specialists to build a multilingual, multimodal AI search engine, marking a significant expansion beyond its chatbot roots. The move intensifies competition with Google and OpenAI whilst the company faces scrutiny in Washington over alleged military connections.
DeepSeek's Search Ambitions
Chinese artificial intelligence startup DeepSeek is recruiting specialists to build a multilingual, multimodal AI search engine, signalling a major expansion of its product offerings as it seeks to compete directly with OpenAI and Google.Multiple job postings released this month reveal plans for search functionality capable of processing text, images, and audio inputs. The company is simultaneously scaling its AI agent capabilities, with job listings detailing requirements for training data, evaluation systems, and dedicated platforms to support agents that operate continuously with minimal human intervention.
AGI Focus and Technical Innovation
The postings, numbering over a dozen, provide fresh insight into DeepSeek's trajectory one year after its R1 reasoning model disrupted the technology industry. In an advertisement for a full-stack developer, DeepSeek sought candidates who possess a persistent curiosity about the technological path and development of artificial general intelligence, underscoring the company's long-term ambitions.DeepSeek recently published a research paper describing a new training method called Manifold-Constrained Hyper-Connections, which analysts described as a breakthrough for scaling large language models. The company has provided limited information about its next-generation flagship model, though a recent reference to model1 appeared on its public GitHub account.
Washington Scrutiny Intensifies
The expansion comes as DeepSeek faces renewed scrutiny in Washington over its connections to China's military. In a letter sent this week to Commerce Secretary Howard Lutnick, Representative John Moolenaar alleged that Nvidia provided extensive technical assistance that enabled DeepSeek to achieve frontier AI capabilities.According to the letter, documents show Nvidia engineers helped DeepSeek optimise algorithms, software, and hardware, allowing the startup to train its models using roughly 2.8 million GPU hours on Nvidia's H800 chips, considerably less than what American developers typically require for comparable systems.
Nvidia pushed back against the allegations, stating it would be illogical for the Chinese military to rely on American technology. China's embassy in Washington criticised the allegations as politicising trade and technology issues.
Published January 29, 2026 at 10:31pm
