gpt-6IMAGE

OpenAI's Astra controllability talk vs DeepSeek's Miku trick

Contrasts OpenAI's serious claims that its Astra model allows unprecedented control of the chain of thought and may prevent x-risk, with DeepSeek's crude trick of telling a model it is Miku so it thinks like Miku.

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)

@teortaxesTexArchived 2026.09.18Originally posted 2026.09.17

OpenAI's Astra controllability talk vs DeepSeek's Miku trick (image 1 of 3)

OpenAI's Astra controllability talk vs DeepSeek's Miku trick (image 2 of 3)

OpenAI's Astra controllability talk vs DeepSeek's Miku trick (image 3 of 3)

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞) @teortaxesTexOriginally posted 2026.09.17

Open original on X

OpenAI: "our breakthrough ASI RSI SSI model, Astra, allows unprecedented controllability of the CoT, we are hopeful this will prevent X-risk, but questions remain – is this deceptive alignment…" DeepSeek: "if you tell ur model that it's Miku it'll think like Miku, lol"

This post quotes another post: x.com/mizorewww/status/2100457470531600764

x.com/teortaxestex/status/2100497885963907558Archived 2026.09.18 10:07