What is DeepSeek?

Axiom Math scored 120/120 on the 2025 Putnam exam, beating the best LLM (DeepSeek at 103/120) and the best human contestant (110/120).

“The best LM deepseek got 103 points out of a 120 point exam. The best human obviously we now know is a student from either MIT or Chicago. We don't know which one because they don't announce the top five winner score got 110 and we got 120.”

GRPO became dominant not due to algorithmic novelty but because DeepSeek did the scaling engineering and shipped a working artifact model that validated it.

“I think the reason GRPO specifically like that algorithm and that acronym like, you know, very concretely took off was not necessarily because it was like a big quantum leap on what came before. It was because DeepSeek did a lot of engineering work around actually scaling it and released an actual artifact model that worked really well with it.”

DeepSeek delivered GPT-4-level performance at 1/600th the inference cost of GPT-4.

“Deep Seek, for example, on GPT-4 was 1/600 the cost.”

DeepSeek demonstrates that natural language verification with meta-verification is sufficient for mathematical reasoning at scale, without requiring formal proof systems like Lean.

“It's interesting that natural language verification with some sort of meta-verification seems to work so far in the published literature.”

DeepSeek uses expert parallelism rather than pipeline parallelism for frontier-scale inference, a deployment choice worth tracking as MoE models proliferate.

“The DeepSeek paper reports what they do, which is that they just do a lot of expert parallelism.”

DeepSeek raised approximately 50 billion RMB (roughly $7.5 billion USD) from Tencent, CATL, and state funds including China's national AI industry investment fund.

“It has raised about 50 billion grand which is like 7.5 USD investors include 10 cent and a few other and oh and also CL the battery maker and some state funds including the national artificial intelligence industry investment fund.”

Nvidia Nemotron Ultra uses Mamba (hybrid) attention, positioning it architecturally differently from Chinese labs such as DeepSeek and MiMo that use sparse attention.

“In the Nvidia Ne Neumitron Ultra used the Mamba attention, whereas, you know, we see, you know, Deep Seek sparse attention and then the MiMo MSA, whatever that stands for, MiMo sparse attention.”

The J-10C dogfight results represent a 'DeepSeek moment' for Chinese military tech, revealing far greater capability than most observers anticipated from a platform that had never before been proven in actual combat.

“Their deepseo or their deepseek moment where Chinese military tech which had never been proven in battle before demonstrated itself to be far more capable than a lot of people anticipated.”

Despite Twitter chatter suggesting GLM is displacing DeepSeek, OpenCode usage data shows DeepSeek dips briefly at GLM launches but then bounces back.

“You can kind of see the DeepSeek one dip around the time GLM sort of comes out, but it's seems to sort of bounce back up after.”

Wall Street analysts are only now updating away from the DeepSeek-era assumption that Chinese labs sell frontier AI intelligence for pennies on the dollar.

“Wall Street analysts and people who aren't necessarily listening to the AI Daily Brief every day are finally beginning to update their priors and recognize that Chinese labs aren't, in most cases, selling frontier intelligence for pennies on the dollar the way they had previously assumed.”