Hacker News Reader

Comment by chrisseaton | Hacker News Reader

If you're hiring someone to do something like implement a new machine learning algorithm or a new compiler optimisation they're not to just pick that stuff up off of stack overflow are they? Sometimes you need specialist knowledge.

View on HN

Most of the time you know when a specialist is needed. Also, the specialist's skill tends to be part of the job description, along with the stuff that is unnecessarily restrictive.

You might come across a ML job advert that says the person needs to know Matlab. Well, what if you're a Numpy or R ML guy? It's quite clear to me that such a person is just as well qualified.

If you're looking for a compiler guy, you want someone who understands those optimisations on a theoretical level, ie abstracted away from the concrete language. The toolset may be completely different for each language, but people with equivalent fundamental knowledge should not be separated just by what tool they happened to use.

If it's a new concept then the entropy is huge for anyone regardless of background.

Do you really think any random programmer can master any given field of computer science in just a few months of starting a job? You could take a compiler engineer and get them doing cryptography, or a cryptographer and get them to write a compiler? No need for any experience to work in either of those fields? Just some JS or Android experience would be fine and you can use Stack Overflow to work out the rest of the details?

Agreed. I think OP's advice only applies to programming languages, and frameworks. But not concepts like machine learning, cryptography, compilers, algorithm design, distributed systems, etc. For those, actual experience does matter.

That's not correct. Who do you think would be able to implement a spam machine learning algorithm in a month? A NLP PhD or a regular Android developer?

Given a whole month? Anyone who can use google and isn't afraid to experiment with the available libraries they find.

Would you trust your security to a TLS implementation done by someone who'd spent a month googling on it?

What if they spent the month finding out how some open source implementation worked, including what experts thought of it, and wired it up for your needs?

You know, at the bottom of your stack someone, somewhere, needs to know a subject well enough to implement things from scratch. If what you are doing hasn't been done before you can't just google or find an open source library. Even if it has been done before often these things are often barely written down.

Do you think when Google decided to create the V8 JavaScript JIT they hired a couple of Django programmers and asked them to google 'how to write a JIT'? Or do you think they specified a certain level of experience in writing JITs on the job advert?

> You know, at the bottom of your stack someone, somewhere, needs to know a subject well enough to implement things from scratch. If what you are doing hasn't been done before you can't just google or find an open source library.

That really depends on what "from scratch" means. Any particular piece of code can be examined in enough detail that nobody in the whole world would understand all of them. Luckily people who understand specific pieces will produce a part for others to use. Often the only thing that hasn't been done before is one particular mashup of existing parts. Then the skill of coding is like the skill of cooking: using judgement to assemble well known parts into a new whole.

> Do you think when Google decided to create the V8 JavaScript JIT they hired a couple of Django programmers and asked them to google 'how to write a JIT'? Or do you think they specified a certain level of experience in writing JITs on the job advert?

That wasn't the point. I'm not saying that any programmer can do any programming job. But that the selection criterion (language, framework) is in general wrong. If you have a guy who did a JIT for Java, do you disqualify that guy from a JIT for Go job? I don't think so.

Still no. Security is hard. Harder than most other stuff

You don't need someone to implement the algorithm, you generally need someone to wire together APIs between your application and machine learning libraries. I'd rather take someone who's been hacking around on Django for a year than a freshly minted NLP PhD for adding spam-detection for messages inside a Django app.

Reply on news.ycombinator.com