Hacker News Reader

Comment by eanzenberg | Hacker News Reader

Super interesting developments..

Also, re:

>AI can't democratize itself (yet?) so I'll help make it more accessible to everyone!

This needs to happen more on the software side. You can buy a world-class quad-gpu machine for about half the cost of a vehicle. You can build a world-class single-gpu machine (what I have) for a fifth of that. It's literally amazing how accessible the hardware is, compared to almost any other science which requires 7 figure funding to get world-class results.

The software, the learning, the math, all need to become more and more accessible so more people can pick up and train an algorithm.

View on HN

I lost access to a TrueCrypt volume from years ago, and using AWS GPU spot instances, I was brute forcing ~1000000 permutations every 40 seconds last night for $7/hr. The hardware is incredibly accessible.

Were you able to crack it?

Yes. Shame on past toomuchtodo for using too few characters, but also thankful I didn't need to spend hundreds of dollars on compute time.

How many characters was your password?

Well, if he was trying 1M combinations every 40 seconds, for $7 per hour, and he didn't need to use hundreds of dollars per hour of commute time, let's say 10 hours = $70. That's 900M combinations per hour, so 9B combinations in 10 hours.

If he was trying combinations using upper-case characters, lower-case characters, numbers, and let's say 20 symbols, that's 82 possible combinations for each one. We'd expect him to find the password after exhausting half of the search set, so we want log base 82 of 18B. That suggests 5 characters.

If he let's say just used lower-case characters and numbers, that's log base 36 of 18B, which suggests 7 characters.

I mean...you can't make math simpler. like...math is math. you can create resources that make it easier to teach math (and for the record high quality textbooks on AI/ML ABOUND). I think you want software and libraries so you can build things without understanding the math. To some degree this already possible (see my english major linkedin friends playing withs sci-kit etc etc). Fundamentally though, I think you need at least some basic understanding of the mathematics to interpret results/create new ideas. Arguing otherwise is like saying you can code without knowing what a queue is...I mean sure you can probably create something that compiles and runs but your horizons will be more narrow than someone with a full computer science education and grounding. just like trying to do ml without at least some understanding of the math is like a blind man trying to open a door.

case-in-point: I am on hn right now because my research is going badly because I am using another researcher's code and I don't fully understand the correspondence between the paper and the code. I am getting terrible results on some multi-class classification (i.e. approximately random). I am not sure if I am using some wrong parameters or because the signal I am looking for doesn't exist in the data. Not knowing at least some mathematics would make this a LOT harder

Re: "I don't fully understand the correspondence between the paper and the code"

Have you tried contacting the paper's author(s)? Preferably the PhD student that did the actual work?

I used to do that when I couldn't make sense of a paper. Not all authors will respond, some might brush you off with a generic remark to read the appendix or whatever. But it's worth a try. In one case, the author frankly told me that the code didn't work as well as it sounded in the paper -- the algorithm was so instable that it wasn't very useful. But they couldn't have written that in the paper, or it might not have been accepted.

Yeah I am collaborating directly with the primary author...but I think you might be right I would have gotten more helpful advice for the graduate student who spent like a year writing this.

Good advice-I'll keep it in mind for future troubles. Thank you!

mean...you can't make math simpler. like...math is math.

Notation makes a huge difference. LinAlg is a good example of this. And a PL is notation.

Simpler notation might make it easier to implement algorithms without fully understanding them. Not sure how much good is going to come from that.

But if you fully understand the problem, notation is not an issue.

I disagree. Try using Roman numerals for example. Or doing linear programming or any sort of optimisation without using matrix notation.

mmm I do see your point. Notation can frame how we think. I hate to be that guy though...but I am a bit triggered by calling linear algebra LinAlg. It is like calling California Cali! Or Analysis -> Anal.

Excellent job mathing, mathperson. May the math be with you!

Reply on news.ycombinator.com