DeepSeek's V3, an open-source model, operates on a significantly smaller budget of $5.6 million, while U.S. firms invest hundreds of millions or even billions for comparable performance.
Andrej Karpathy remarked on DeepSeek's V3 as 'a highly impressive display of research and engineering under resource constraints,' highlighting its efficiency despite limited resources.
DeepSeek's CEO Liang Wenfeng emphasized that financial resources are not the core problem for them; rather, restrictions on advanced chip shipments severely hinder their development.
Despite U.S. export restrictions aiming to curb China's advanced AI growth, such measures seem to have inadvertently spurred Chinese innovations in building efficient AI models.
Collection
[
|
...
]