Adversarial Examples Are Not Bugs, They Are Features · HackerPapers discussions

Adversarial Examples Are Not Bugs, They Are Features — mentioned by dijksterhuis

Mon, 27 Apr 2026 12:05:19 GMT

i’ve caught myself doing the “it’s this and that — it’s not the other” thing a few times. i dunno if it’s because i’ve seen it so many times because of AI generated comments etc and that’s become a norm in my brain, or if it was actually something i do regularly and ive just never noticed it. it might be the latter, because i always got the title of this paper https://arxiv.org/abs/1905.02175 backwards. i used to write it adversarial examples are features, not bugs (which is apparently not correct in english language 0_o) regardless, ive started editing it out when i notice ive done it now.

Paper and discussion history. A mention is not an endorsement.

Adversarial Examples Are Not Bugs, They Are Features — mentioned by b3kart

Wed, 05 Feb 2020 17:27:15 GMT

I am not convinced the explanations we’d get from these models would be interpretable enough to drive their conceptual improvement. Deep nets just don’t learn nice interpretable features, nothing in the training objective makes them do so. E.g. they make use of really dubious features [1], and might not actually learn “hierarchies” of features as we thought they were [2]. [1] https://arxiv.org/abs/1905.02175 [2] https://arxiv.org/abs/1904.00760

Paper and discussion history. A mention is not an endorsement.

Adversarial Examples Are Not Bugs, They Are Features — HN submission

Wed, 08 May 2019 23:23:35 GMT

Adversarial Examples Are Not Bugs, They Are Features

76 points; 28 discussion replies on HN. Paper and source history. A submission is not an endorsement.