Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)(@teortaxesTex):They objectively have stronger models and they are "releasable" in the sense of being used internally at scale. The private-public capability gap used to be 6 months or so in pre-GPT-4 era. By 2025 it had shrunk to maybe 3 months. Now it's growing again, I believe.

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex

2026.11.10 14:40

They objectively have stronger models and they are "releasable" in the sense of being used internally at scale. The private-public capability gap used to be 6 months or so in pre-GPT-4 era. By 2025 it had shrunk to maybe 3 months. Now it's growing again, I believe.

Bora Tas@Tas1Bora

2026.11.10 01:24

Labs internally having models that are releasable and much ahead of the public ones is not realistic. They are bleeding money like crazy. They wouldn't hold technology they have back. I liken this to "Apple has iphones of the future ready" claims from the early-10s.

コミュニティへ転送