Adaptive Learning at the Error Threshold