Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Companies from the YC Summer 2018 Batch

blog.ycombinator.com

Post reply on HN
Print view View original

31–40 of 99 posts

Re: Companies from the YC Summer 2018 Batch

#31

Post by abhisuri97 »

I found CSPA to be an interesting proposal. I'm currently studying CS and Biology and found that the comparison of the exam to the "SAT" of computer science is a bit of a misnomer. The SAT requires minimal prior knowledge to take (albeit if you want to score well, you end up needing to study for it through test prep books etc). I think perhaps it may be more likened to an MCAT equivalent (or any professional school exam equivalent) because of the need to know about certain subject fields.

Someone also brought up a point regarding the validity of testing for skills in web development if test takers won't end up in those fields anyway. However, when looking at similar exams (e.g. MCAT), you can make the argument that the majority of doctors won't use Organic Chemistry or won't need to know about the minutiae of Psychology & Sociology to be successful. The MCAT becomes a successful discriminator of good vs bad test takers because of the amount of work and critical thinking skills that is needed to cover the litany of subjects tested. Similarly, the CSPA could follow a similar pattern as the MCAT by testing a range of subjects as a means of measuring one's ability to learn a large sample of topics.

The only thing I can really think of that CSPA can improve upon is providing more preparatory materials. Additionally, it'd be useful to see how people end up performing on the exam and if there is an actual "Bell Curve" like distribution of scores that results.

Re: Companies from the YC Summer 2018 Batch

#32

Post by ethbro »

reikonomusha wrote: post #14
phkahler wrote: post #4

Personalize testimonials to the person reading them? Did I read that correctly? Fake reviews taylored to the reader?

I take it as: “See the reviews which will skew your perception positively.” Doesn’t sound enticing as a consumer. I just want the facts.

Devil's advocate. A ML service offering being viewed by two users -- an engineer and an executive.

Even with 100% truthful reviews, they're interested in different things and need very different language.

Re: Companies from the YC Summer 2018 Batch

#33

Post by cspa »

abhisuri97 wrote: post #31

I found CSPA to be an interesting proposal. I'm currently studying CS and Biology and found that the comparison of the exam to the "SAT" of computer science is a bit of a misnomer. The SAT requires minimal prior knowledge to take (albeit if you want to score well, you end up needing to study for it through test prep books etc). I think perhaps it may be more likened to an MCAT equivalent (or any professional school e…

Great feedback! We do have a distribution of scores for our latest exam session here: https://medium.com/@cspa_exam/july-2018-cspa-results-509ace9...

Re: Companies from the YC Summer 2018 Batch

#34

Post by mmt »

reikonomusha wrote: post #10

CSPA [1] sounded interesting, and is something similar to what I’ve been wanting to do, but after looking at the sample exam, I was pretty disappointed. I thought it would be about computer science and engineering fundamentals. It has some of that (e.g., representation of integers, memory access speeds, etc.) but I also noticed it’s full of completely inessential things, like escaping XHTML, identifying valid JSON pa…

Their topics page https://cspa.io/topics also has numerous duplicate "example" links. Some of them are barely related to the sub-topic line item and some not at all (perhaps a typo? copy-pasto?).

I provided feedback that included this issue, I believe around when it was first a "Show HN" post, but the page was never updated.

That would certainly give me pause as to the accuracy of a technical (or detail-orientation) assessment.

Re: Companies from the YC Summer 2018 Batch

#35

Post by cspa »

ABCLAW wrote: post #30
cspa wrote: post #28

Earlier quoted context omitted.

Excellent points -- this is something we discuss extensively. After talking with so many employers, we found that each of them look for something different. Adaptive scoring algorithms and ways to let employers create custom projections of our multidimensional data is something we're thinking about. If you have ideas on how to build this, I'd love to hear them! james@cspa.io

Hey James. The issue you're having is that you're trying to test over 7 dimensions, with the majority of those dimensions being irrelevant to an individual taking the test to signal mastery of 2-3 of them. The standard method in testing to deal with issues of this nature is to provide people with the opportunity to test themselves on modules, rather than taking the omnibus test so that they can spend the majority of…

Re: custom modules. One (arguably) interesting thing we do is to only take the MAXIMUM subscore of all 7 topics and use that component in the composite score, ignoring the other 6 topics. This allows candidates to focus on only certain topics if they want. Employers want someone who is strong in one area, rather than a jack-of-all trades and master of none.

Adaptive testing (https://en.wikipedia.org/wiki/Computerized_adaptive_testing) is something we're looking into, and would be extremely interesting (and challenging!) to build.

Re: Companies from the YC Summer 2018 Batch

#36

Post by AriaMinaei »

Optic [0] caught my eye. They seem to have made possible a new kind of abstraction in code. Something that's different from both functions and macros. It's a kind of abstraction that can be arbitrarily customized at each call site, yet retain its identity across all of them.

Example:

  foo = do_a()
  bar = do_b(foo)
Example 2:

  foo = do_a()
  bar = do_z(foo)
  baz = do_b(bar)
Example 3:

  foo = do_a()
  bar = foo + 10
  baz = do_b(bar)
The three examples have strong similarity in their first and last statements. There is a pattern there, but that pattern cannot be abstracted into a single macro or a function. So this pattern does exist, and is recognizable to the human eye, but the language does not allow one to express it.

What Optic seems to do is to recognize the pattern, create a single model out of it, and allow you to re-use that pattern elsewhere, or even transform it into new ones and re-use those new patterns elsewhere.

Again, this would be a new kind of abstraction. One whose leaks are easier to fix. You can have your cake and eat it too!

[0] https://useoptic.com

Re: Companies from the YC Summer 2018 Batch

#37

Post by ABCLAW »

cspa wrote: post #35
ABCLAW wrote: post #30

Earlier quoted context omitted.

Hey James. The issue you're having is that you're trying to test over 7 dimensions, with the majority of those dimensions being irrelevant to an individual taking the test to signal mastery of 2-3 of them. The standard method in testing to deal with issues of this nature is to provide people with the opportunity to test themselves on modules, rather than taking the omnibus test so that they can spend the majority of…

Re: custom modules. One (arguably) interesting thing we do is to only take the MAXIMUM subscore of all 7 topics and use that component in the composite score, ignoring the other 6 topics. This allows candidates to focus on only certain topics if they want. Employers want someone who is strong in one area, rather than a jack-of-all trades and master of none. Adaptive testing ( https://en.wikipedia.org/wiki/Computerize…

Hi James,

The algo to take Core + the best other section, right? As it stands, the test difficulty makes it pretty easy to hit 2 perfect scores, so you'll be competing against those pretty consistently.

How does the CSPA help my interviewing process if I'm a network guru and knock all those questions out of the park, but apply for a ML job where my knowledge is paltry? How does my CSPA testing help me differentiate from the above guy as a network guru when I also aced the ML section? This seems like an obvious area the test can outperform other testing metrics.

As far as I can tell, the only value for the current scoring system is to weed out people who are just plain abysmal at everything. Eventually people are going to game your exam and pop up very close analogues onto google. At that point even the floor function is dead with the added deadweight cost of the exam still being an application requirement at BigCo.

Maybe I'm just blinded because I think having accurate skill radar charts that test takers and employers could use for self improvement and prospect evaluation, respectively, could be an absurdly large value add.

Re: Companies from the YC Summer 2018 Batch

#38

Post by goatherders »

maxxxxx wrote: post #8
chiefofgxbxl wrote: post #6

Cool batch, particularly LemonBox which provides personalized vitamin packs to buyers in China. There's a huge user base right there, and I imagine the U.S. market is largely saturated / competitive already. I must respectfully express my concern over: "Grabb-it Inc. turns rideshare cars into digital billboards." There's probably a market for it as online advertisers face an uncertain future with possible regulation,…

Totally agree. Coming up with an idea to serve even more ads is not something to be proud of. Take your money and develop some new technology instead.

Why? This hits at the heart of startup porn. It's about money. It's only about money. It's only ever been about money. This is a good way to make money so it exists. I'm happy for startups that don't hide behind their capitalist mission.

Re: Companies from the YC Summer 2018 Batch

#39

Post by maxxxxx »

goatherders wrote: post #38
maxxxxx wrote: post #8

Earlier quoted context omitted.

Totally agree. Coming up with an idea to serve even more ads is not something to be proud of. Take your money and develop some new technology instead.

Why? This hits at the heart of startup porn. It's about money. It's only about money. It's only ever been about money. This is a good way to make money so it exists. I'm happy for startups that don't hide behind their capitalist mission.

True. I just don't think we should celebrate these startups.

Re: Companies from the YC Summer 2018 Batch

#40

Post by cspa »

ABCLAW wrote: post #37
cspa wrote: post #35

Earlier quoted context omitted.

Re: custom modules. One (arguably) interesting thing we do is to only take the MAXIMUM subscore of all 7 topics and use that component in the composite score, ignoring the other 6 topics. This allows candidates to focus on only certain topics if they want. Employers want someone who is strong in one area, rather than a jack-of-all trades and master of none. Adaptive testing ( https://en.wikipedia.org/wiki/Computerize…

Hi James, The algo to take Core + the best other section, right? As it stands, the test difficulty makes it pretty easy to hit 2 perfect scores, so you'll be competing against those pretty consistently. How does the CSPA help my interviewing process if I'm a network guru and knock all those questions out of the park, but apply for a ML job where my knowledge is paltry? How does my CSPA testing help me differentiate f…

Yep I agree! In fact, about 50% of the people who take the CSPA, do it more for self-evaluation than for employment purposes.

You're right, so far most of the test takers are entry-level or changing careers. To accurately assess specialities like ops/IT, we'll need separate, dedicated subject tests (or adaptive tests).

That said, no one has yet gotten a perfect 1600 :) . The highest is something like 1380.

Post reply on HN