-
open source
2026-07-22 ~
↺ 立場変更
Reaffirming open-weights commitment: the open-sourced model is identical to the one they deploy themselves.
«Is the open-source model we give the same as the one we deploy? Yes, it is the same.»
-
products
2026-07-22 ~
Restraint strategy: consumer and enterprise products are mere byproducts on the road to AGI.
«Consumer and enterprise products are byproducts on our road to AGI.»
-
agi timeline
2026-07-22 ~
New framing of the AGI path: the core capability of the next-gen model must be continuous learning.
«The core capability of a next-generation model must be continuous learning.»
-
china us
2026-07-22 ~
Quantified efficiency: about two years behind the US, but with one-twentieth of their compute.
«Two years behind the US, and did it with only one-twentieth of their compute.»
-
compute
2026-07-22 ~
Domestic hardware: Huawei's 950 supernode can fully substitute Nvidia's GB200/GB300 on performance and price.
«Huawei's 950 supernode can fully substitute Nvidia's GB200 and GB300 on performance and price.»
-
compute
2026-07-22 ~
↺ 立場変更
Shift on chips: Nvidia's CUDA moat is being rapidly dismantled.
«Nvidia's CUDA moat is being rapidly dismantled.»
-
agi timeline
2024-07-17 ~
AGI timeline: maybe 2, 5, or 10 years, but it will happen within this generation's lifetime.
«AGI could be 2, 5, or 10 years; in any case within our lifetimes.»
-
compute
2024-07-17 ~
Money was never DeepSeek's problem; the real constraint is the ban on high-end chips.
«Money has never been our problem; the embargoed high-end chips are the problem.»
-
open source
2024-07-17 ~
In disruptive tech a closed moat is short-lived; open-sourcing and publishing cost them nothing.
«A closed-source moat is short-lived. Open-sourcing and publishing papers — we lose nothing.»
-
jobs
2024-07-17
The breakthrough V2 team is entirely local talent, no overseas returnees.
«The V2 model has no one who returned from overseas; they are all local.»
-
china us
2024-07-17
The real China-US gap is not a year or two but originality vs imitation.
«The real gap is original vs imitation. If this doesn't change, China will only ever be a follower.»
-
other
2023-05-23
On innovation: by nature expensive and inefficient, sometimes wasteful — justifying research over imitation.
«Innovation is expensive and inefficient, sometimes accompanied by waste.»
-
open source
2023-05-23
Wants large models cheaply usable by many, down to small apps, not held by a few companies.
«We hope more people, even a small app, can use large models cheaply, not tech held by a few.»
-
agi timeline
2023-05-23
The company's goal is not applications but building AGI; he sees LLMs as the path to it.
«What we want to build is AGI. Large language models may be the necessary path to AGI.»
