Amazon CEO Andy Jassy, impressed by DeepSeek, says “we thought that was…”

Amazon CEO Andy Jassy has heaped praise on DeepSeek, saying that the company was specifically impressed with some of the training techniques that the Chinese AI startup has used to train its artificial intelligence (AI) model.
“Well, I’d say a few things because there are a few questions built into that. First of all, I think like many others, we were impressed with what DeepSeek has done,” Jassy said while replying to a question during Amazon Q4 earnings call.
“I think in part impressed with some of the training techniques, primarily in flipping the sequencing of reinforcement training — reinforcement learning being earlier and without the human-in-the-loop. We thought that was interesting ahead of the supervised fine-tuning,” he added.
Jassy also highlighted “some of the inference optimisations” that DeepSeek did, saying they “were also quite interesting.”

DeepSeek available on Amazon Bedrock

Jassy also said that DeepSeek’s AI model is available on Amazon Bedrock.
“And I think if you run a business like AWS and you have a core belief like we do, that virtually all the big generative AI apps are going to use multiple model types, and different customers are going to use different models for different types of workloads. You’re going to provide as many leading frontier models as possible for customers to choose from. That’s what we’ve done with services like Amazon Bedrock. And it’s why we moved so quickly to make sure that DeepSeek was available both in Bedrock and in SageMaker faster than you saw from others,” he added.
He also believes the cost of inference will come down which will make it easier for companies to be able to infuse all their applications with inference and with generative AI.