DeepSeek or DeepFake? Our vultures circle China's hottest AI
If this keeps Silicon Valley on its toes and honest, who are we to complain?
Kettle There's really only one topic for the Kettle this week. DeepSeek.
What began as a Chinese hedge-fund venture has blown away nearly a trillion dollars in stock market value from Nvidia, Microsoft, and Meta. But are DeepSeek's freely available V3 and just-released R1 LLMs all they are cracked up to be? We have our doubts, as we discussed below for this week's episode.
For one thing, DeepSeek argued training V3 in the cloud would cost less than $6 million, a figure arrived at by multiplying the 2.78 million Nvidia H800 GPU hours it apparently took to train the LLM with a cloud GPU rental rate of $2 an hour. Yet it's clear the lab has spent many millions on thousands of its own GPUs and research and development costs to build its models in-house; one analysis reckons the Chinese lab has spent $1.6 billion on hardware.
On the other hand, DeepSeek claims its architecture is efficient, and doesn't appear to need the zillions of dollars in compute investment talked up by Silicon Valley.
Joining us today are Tim Prickett-Morgan of The Next Platform, Tobias Mann and Iain Thomson of The Register, and editor-in-chief Chris Williams.
Tim earlier dived into the backend tech at DeepSeek here, and Tobias described how to run the DeepSeek R1 locally over here. Our 20-minute catch-up may contain technical assessments and derisory sarcasm.
- DeepSeek means companies need to consider AI investment more carefully
- DeepSeek stirs intrigue and doubt across the tech world
- DeepSeek's not the only Chinese LLM maker OpenAI and pals have to worry about. Right, Alibaba?
- Guess who left a database wide open, exposing chat logs, API keys, and more? Yup, DeepSeek
The Kettle is also available audio-only via RSS and MP3, Apple, Amazon, and Spotify. Thanks for listening! ®
Biting the hand that feeds IT
