OpenAI's 10% safety compute questioned for GPT-6.1 Astra alignment

Token Gremlin @TokenGremlinOriginally posted 2026.10.18

Open original on X

OpenAI has allocated 10% of its total compute (and that's A LOT!) to its weakest area: safety. If alignment was the problem with GPT-6.1 Astra or any other model, that much compute should be enough to fix it FAST and deliver a real answer to Opus 5.5 in a month, not a year.

This post quotes another post: x.com/polydaic/status/2111834405618561098