Closed Gurkenglas closed 1 year ago
The logit lens describes a supisciously legible pattern in GPT's internals. Let's play around with it: Can we enforce it? Can we suppress it? Call Gurkenglas on discord for rambling.
The logit lens describes a supisciously legible pattern in GPT's internals. Let's play around with it: Can we enforce it? Can we suppress it? Call Gurkenglas on discord for rambling.