musing

[ajva]

Random pensification sparked by a comment from josh_the_cat in adjectivemarcus's journal.

It's really annoying when people go on and on about a subject you know a lot about, and they quite clearly know very little and are wrong about many things, but you have to listen to them because 'everyone is entitled to their opinion'. I had this experience yesterday, when I was in a debate with someone about the implications of the BBC threatening to pull its channels from BSkyB. Now, this is my job. But they didn't see that as a reason that I might be any more right than them. It used to piss me off no end also when people would give advice about music and the band. Fucking endlessly. Everyone's pet subject, you see. 'You should send a demo tape to record companies'. 'You should dress like X'. 'You should play a festival'. 'You need to sharpen your sound'. I really used to hate that.

Comments

[conflux.livejournal.com]

I have a horrible feeling that I may have once mentioned something like this to you. I seem to remember wondering out loud to you if neural network techniques could be usefully applied to encryption. In my defence I do know a lot about neural networks, if rather less about encryption. Some early self learning networks were based on code-book theory which was originally developed for combined encryption and compression applications, or so I was lead to understand by a certain very famous Hungarian prof.

Oh yes, and my job has recently involved me implementing an encryption scheme that uses SSL but I just plugged in existing code - I didn't develop any of the actual encryption algorithms myself.

Mind you, from my experience at Cambridge, sometimes if you actually want an expert in a field to explain something to you, the best way to get them to do it is to show how clueless you are so that they can have fun correcting you. Also, from the other side, if you don't let clueless person talk first then you don't know where the gaps in their knowledge lie and can spend ages explaining something they already know.

[ciphergoth.livejournal.com]

If it's that early in the history of self-learning networks, the work you're talking about may predate the very existence of an open cryptographic scientific community, which first started to come about in the mid-1970s.

And I'm all in favour of the learner speaking first, so long as it's not to tell me about their great new idea. The worst of it is that it's usually a very awkward, inconvenient, slow, hard-to-analyse solution to symmetric encryption, a problem for which we already have very clean, convenient, fast and straightforward to analyse solutions. I don't use the word "insecure" because they are usually "not even insecure" in the same way that it's better to be wrong than "not even wrong".