DeepSeek has drawn comments from everyone ranging from US President Donald Trump to Elon Musk, and Google too has now commented on the model.
Demis Hassabis, the CEO of Google DeepMind, has said that while DeepSeek is an impressive model, the Chinese company hasn’t actually invented anything novel. “It’s a very impressive model, very impressive piece of work,” Hassabis said in an interview.” I think the team is probably the best team that I’ve seen come out of China,” he said.
But Hassabis said that not everything that was in the public domain about the model was necessarily true. “So that said, I think a lot of the claims are exaggerated and a little bit misleading. First of all, when you report how much it costs to do a training run, they seem to report just their final training run, which is actually a fraction of the cost it normally takes to explore and train and do all the tests beforehand before you do your final run. So we think that’s been sort of underestimated,” he said.
“And then the other thing is they seem to have relied on some western models to distill from or to basically fine tune against the outputs of. You have to sort of factor that cost in as well. And then finally, like, it’s an impressive piece of work, but we don’t see any silver bullet new technologies that or techniques that we haven’t seen before, haven’t invented before. They just applied it very well,” Hassabis continued.
“So it’s impressive, but it isn’t some new outlier on the efficiency curve. For example, Gemini is more efficient than Deep Seek in terms of its training to performance or its cost to performance. We just don’t talk about that it very much, but it’s actually more efficient on the Pareto frontier of cost to performance,” he claimed.
DeepSeek has drawn similar claims from other US companies. Anthropic CEO Dario Amodei had said that while DeepSeek was being marketed as cheap, it had more than 50,000 NVIDIA GPUs. US President Donald Trump had said that DeepSeek was a wake-up call for the US AI industry, and Elon Musk had said that his company would be soon releasing AI models superior to DeepSeek. And with Google too now saying that they already have models better than DeepSeek, it’s clear that while the tech industry in the US acknowledges the advancements made by DeepSeek, they don’t seem to consider them to be something that US firms can’t reproduce.