-
compute
2026-03-18 ~
NVIDIA GTC talk: token efficiency is not about saving cost but about raising the ceiling of intelligence.
«Token efficiency is not just about efficiency. It is about raising the ceiling of intelligence itself.»
-
agi timeline
2026-01-29
↺ स्थिति बदली
Emphasis shift: intelligence ceiling now hinges on new learning algorithms (earlier prioritized scale).
«The ceiling of intelligence depends more on whether we can invent new learning algorithms.»
-
products
2026-01-29
Teases K3 scale: confident K3 will be much stronger than K2.5, even if not 10x.
«I believe Kimi K3, even if not 10x stronger than K2.5, will be much stronger.»
-
compute
2026-01-29
On the GPU gap with the US: says the gap has not narrowed.
«I don't think the gap has narrowed.»
-
china us
2025-11-11 ~
On competing with OpenAI/Altman: Moonshot has its own way and pace.
«I don't know, maybe only Sam knows; we have our own way and pace.»
-
compute
2025-11-11 ~
On compute under export limits: trains on H800, admits a disadvantage in tier and quantity, but maxes out every card.
«We use H800 GPUs with Infiniband... but we made full use of every single card!»
-
open source
2025-11-11 ~
↺ स्थिति बदली
POSITION SHIFT: after open K2/K2 Thinking, embraces open source — AGI should lead to unity, not division.
«We embrace open source because AGI should be a pursuit toward unity rather than division.»
-
compute
2024-03-04
Scaling-first principle: if scale solves it, don't invent a new algorithm; an algorithm's value is enabling better scaling.
«If scale can solve a problem, don't use a new algorithm; a new algorithm's value is enabling better scaling.»
-
agi timeline
2024-03-04
Long horizon: AI is not about finding PMF in a year or two, but how it changes the world over 10-20 years.
«AI isn't finding a PMF in the next year or two, but how it changes the world in 10-20 years.»
-
china us
2024-02-21 ~
Believes AGI will be inherently global, not national.
«I firmly believe AGI will be inherently global.»
-
products
2024-02-21 ~
Long context as the path to AGI: with a billion-token context, today's problems cease to exist.
«If you have a context length of 1 billion, today's problems will cease to be problems.»
-
agi timeline
2024-02-21 ~
States his core thesis: AI is fundamentally a stack of scaling laws.
«AI is fundamentally a pile of scaling laws.»
-
open source
2024-02-01 ~
Early stance: skeptical of open source, says it is usually laggards who open-source.
«it's usually the laggards who might do that»
-
products
2023-10-12
At Kimi's 200k-character launch, argues only in-house models create UX differentiation.
«Only self-developed models can create differentiation in user experience.»
