Is the new model from DeepSeek here?
Wallstreetcn·- DeepSeek is advancing its new version model, with a greyscale test before the official V4 launch, featuring an expanded context length from 128K to 1M. - Nomura Securities reports that the upcoming V4 model aims to drive AI commercialization through innovative architecture, significantly lowering training and inference costs. - This shift is expected to stimulate demand for AI applications and reshape the competitive landscape, benefiting Chinese AI hardware manufacturers while software companies might see enhanced value.No Business Model - DeepSeek's Strongest "Moat"
Wallstreetcn·Liang Wenfeng's new paper: Provide a "dictionary" for large models, with computation and memory separated, IQ skyrockets, spoilers for DeepSeek V4?
Wallstreetcn·- DeepSeek's research presents a new "Engram" module that breaks the expensive paradox of the Transformer architecture by separating memory from computation in AI models. - This separation allows static knowledge to be stored efficiently while freeing up model capacity for more complex reasoning tasks, leading to unexpected improvements in logic and coding capabilities. - The upcoming DeepSeek V4 model, set to release before the Lunar New Year, aims to capitalize on these advancements, potentially redefining AI standards by enhancing memory capacity and inference efficiency.