Company · mentioned 120 times
Chinese AI company that develops large language models and open-weight models, known for architecture innovations such as multi-head latent attention.
Startup Insider· 9 Nov
Daniel Höpfner says DeepSeek is valued at 75 billion.Training Data· 5 Nov
Ronak Malde says DeepSeek V4.1 Flash, which came out recently, changed much of the base transformer architecture, using fast and slow memory within the weights and skip connections between layer 2 and layer 50.Latent Space: The AI Engineer Podcast· 4 Nov
Shawn Wang says manifold hyperconnections are what DeepSeek does.Startup Insider· 4 Nov
Jan Thomas says there are rumours that DeepSeek had only 6 million dollars available for chip power, but that he does not know whether that is true.Property AI Report· 2 Nov
Mal McCallion says Nvidia beat Google's fall once, when DeepSeek first popped out.