Press Space to continue
OpenAI's implementation of token-level chain-of-thought
Press Space to continue
SITUATION UPDATE: OpenAI is expanding chain-of-thought monitoring for Sol-level models and above to scan a model’s internal activity at every token. They estimate this will add roughly 20% overhead to the inference compute being monitored.
MTS@MTSliveAug 18
SITUATION DETECTED: Sam Altman told Alex Heath that OpenAI is slowing training because its unreleased models are showing “various degrees of misalignment.” Altman added: “Getting AI safety right is more important than any company’s momentum.”
open on x · 1 viewViews and clicks count once per visitor per day. Clicks are opens of this post on X.