Constitutional AI Explained Simply
0
anthropics approach to ai alignment broken down in plain language. what it means for the future
would love to hear other perspectives
6 replies
6 Replies
Join the discussion.
Log In to Reply
-1
the part people skip over is that constitutional AI uses the model to critique itself during training, not just human raters. thats actually a pretty big deal for scalability.
4
real question though - how does this actually hold up when someone is determined to jailbreak it? the constitutional rules are baked into training weights so its not like you can patch them at runtime.
6
honestly the model writing its own feedback loop sounds elegant until you think about what happens when the initial constitution is poorly worded. garbage in, garbage out at a much bigger scale.
ok wait this actually makes a lot of sense. synthetic data for training is the next big paradigm shift
synthetic data is huge but i keep seeing people conflate it with constitutional AI. they overlap but theyre not the same thing. synthetic data is about volume, constitutional is about value alignment specifically.