deepGem

I wrote a couple of these home works and finally gave up. In fact, I won't interview for any dev positions that require me to do a recruiter phone screen, another technical phone screen, a homework and then an onsite coding test. Expecting production level code in a homework is just asinine.

I'd much rather do a timed coding test for 30 to 60 minutes and be done with it.

If you are still on this thread - check this video out at 28:56

https://www.youtube.com/watch?v=4inIBmY8dQI&feature=youtu.be

One of the issues that I faced during my short stint building ML models for fraud detection in debit card transactions was dealing with class imbalance. I was not completely convinced that over sampling techniques or under sampling techniques would work. My initial experiments just resulted in more false positives. Just curious if you guys faced similar problems.

The other point I bring about is rather rhetorical - There are no open standards, model baselines and datasets in the Fraud domain. Compare building a model for fraud detection to building a model for image recognition or object detection There is a standard baseline, standard datasets and your model competes against that baseline. Because of the open nature of image recognition, the models have improved astronomically. I feel that a lack of such openness is fraud is holding back on innovation. I could be wrong in this assessment so please correct me if so.

Yep, I agree but such an obsessive focus comes at a cost of execution speed. In the early stages I'd rather be fast and a bit sloppy than be slow and perfect.

Sadly no, I tried finding it everywhere but no luck, should have taken a backup of our emails before deleting our domain mail boxes.

We applied to YC once and were rejected. Didn't take it personally but since we were in a very early stage, we dropped the idea.

I've personally found the YC application extremely gratifying to write. It does help you answer those key difficult questions that you'd have otherwise glossed over because of complacency or over confidence. What I've thought of doing is to fill the YC application for the next idea that I get and keep refining it as I build upon the idea. Kinda use the application as a good idea refining template. If at any stage you feel that you are confident in all the questions, you can either apply - literally no cost or just keep building. Having solid answers to all the questions - means that your idea has some potential of turning into something big. Of course, your idea could turn out to be a better business even without having all the answers. What I've experienced is that as you write the answers, you'll feel really confident if you are doing the right thing.

Another aspect in the application that I had a conflict with my co-founder at the time was the amount of time to spend on the application. He was anal about getting every aspect of the application right. We shot the video like a 100 times. I contented that was unnecessary but eventually agreed to disagree. I feel that if you are on the right track, you shouldn't spend more than a couple of hours on the final application, regardless of the stage you are applying at.

It takes some practice to get the video right. This again depends on the co-founder dynamics. I would use the video prep as a proxy to estimate the efficacy of the co-founder dynamics. If you took really long to nail the video - I'd spend some time introspecting. All said and done, if you take a really long time just to nail a video, how would you fare in dealing with more complicated company decisions.

So IMHO there is a lot of value in writing the YC app, regardless of whether you make it or not.

Very true but somehow Wall St. is valuing Tesla very differently than they valued Amazon 15 years ago. Tesla's stock hasn't gone down that much, there's volatility but not the near 100% crash that struck Amazon. On the flip side, Tesla has needed a lot more capital than Amazon and they have issued a lot more shares to raise that capital (close to 12B now). I don't think Tesla's stock price, despite it's volatility reflects it's debt economics.

I think the phrase 'Auto-pilot' is a misnomer. It's a lead-on phrase, confuses people to let their eyes off the road. They should have named it 'driver assist' or something mundane people shouldn't get confused by. Auto-pilot for any person means that the car will steer automatically, which it does, but also kills you. At least by now, it should be obvious that a half baked auto-steering or auto-pilot is a colossal mistake. Admit that you made a mistake and be done with it.

Tell people explicitly not to use auto pilot. Tesla can still use the onboard cameras and everything else for training the AI.

I don't have a current website yet. Thanks for pointing this out. I have a static page at callindra.com (works without the www) I have to fix the Godaddy DNS. Will do.

I started working on recording meetings and converting the recordings to meeting minutes with actionable items. Not a trivial task. My idea was to build an iPhone app with ML in the cloud. I did some initial market survey and found no takers for this idea. I spoke to startups, mid-enterprises and even law firms. I got one positive response from a VC who said this could be useful in tracking the meetings he has with founders as most of them go untracked but I was not too convinced (perhaps I was wrong). So dropped that idea and tried on some ideas around computer vision, prompted by the problems some of the customers were elucidating to while validating the above idea. "We don't really care about meetings but can you use machine learning to monitor people ? I am not a big fan of surveillance. After this I have drifted quite a bit around the last year or so. Current situation - no direction. Just having a side thought on doing something around high definition maps. Looked for a job to fill up some savings, but boy the interviewing phenomenon is nuts. I haven't interviewed in like 15 years. So got back to doing leetcode and Hackerrank.

"Maybe there's a small but reasonably profitable endeavour out there" I am actually thinking about just this. Let's see.

I more or less left full time work when I was 39 to pursue startups. The reasoning was exactly on the lines of what you have described. It's been 2 years and the journey has been fraught with failures. One, it's become really hard for me to find a co-founder. Almost none of the guys in my friend circle are willing to take the risks associated with startups. I've attended a pre-accelerator to find a co-founders but haven't had any luck in finding the right person. Two, I'm just getting lazy because of the financial cushion. I mean, it's not much but still some sort of lethargy has set in since I don't have a pressing problem to solve. Until I get to the stage where things kick off, I bet this is going to be the norm. On one hand I have a financial cushion, on the other I know that the cushion won't mean much in the next 10 years. So, I'm constantly thinking about self sufficiency down the line. So, a combination of no cofounder + no exponential idea + procrastination = My life has become quite miserable. I mean, sure on the surface I project an image of it's all good, if you know what I mean.

Well, most software engineers don't despise managers. Engineers despise suits who don't understand software engineering. Sadly, most suits fall into this category and to make matters worse they turn out to be control freaks. They want to control something that they don't understand and this pisses engineers off. Managing a team of talented software engineers is more art than science. Good luck trying to win the hearts of engineers while being a control freak who doesn't understand what the team is doing.

All the prominent CEOs and execs who stand out are exceptions to this case. They foster strong team building and give complete freedom for the team to execute. They own up to their mistakes. Most of them are not control freaks, but they are detail oriented (Jobs). It's a joy to work with such managers.

"The data yield evidence that rides from commercial banks directly to the New York Fed and offsite meetings involving insiders of the New York Fed and commercial banks increased around the time of FOMC meetings"

Well, wait a minute. The data is anonymized so how is the author inferring that the trips were taken by insiders ? It could be anyone dropping off at the Federal reserve. There is high likelihood that these are insiders. Did the same insiders have any say in the meetings, were they even present in the meetings is hard to determine.

What can be safely inferred from this analysis is that there is an explosion of physical interactions between the NY Fed officials and the banks once the embargo is lifted. Nothing wrong with that I suppose.

Great! Thanks Chris. Look forward to the white paper.

Very interesting. They are baking auto differentiation/gradient computation directly into the compiler ? How will this be work, since you invoke TF and TF computes gradients (at least on the loss functions) using Python/C++ from first principles.

It'll be nice to use the Xcode build system, but given that Xcode isn't open source, I wonder how this will play out. Is there a way to use Bazel natively with Swift ?

Exciting times though !

Yeah, the 'best' care is needed mostly for critical life threatening situations and for emergency situations such as accidents/fire etc. If one were to see the relief camps or first aid offered during war, it becomes pretty evident that a well sterilized plastic bag should suffice for IV. Probably not even costing 50$.

Isn't this primarily on Android ? Am quite certain no app can monitor your location in the background on iOS.

Regardless What pisses me off is FB not providing evidence to the contrary, that they don't harvest data in the bqckground. Not providing this evidence makes it quite clear that they do harvest this data. The issue isn't about permissions. 'Sir, may I come inside your house, is fine, but ransacking the house once you get is is douchebaggery.

I concur, as an immigrant who lived in NYC for a couple of months, not once I felt like a foreigner. Some other cities in the US where I've absolutely never felt like a foreigner - Bay area, Portland, Seattle and Vancouver BC. Surprisingly, I felt like a foreigner in London, which is the most culturally diverse city and also in Singapore.

How do you mean ? From what I've observe all fake reviews tend to fall into a pattern. Usually a mix of broken sentences and hyperbole. I find the positive fake review patterns far simpler to detect than the negative review patterns. Nonetheless, a decent sentiment classifier can be modified and used to classify fake reviews, the fake positive ones at least.

Starting strength is a good book to learn about barbell squats. Pretty simple to understand and the postures are mostly based on angles w.r.t joints.

Anecdotal reply, dedicated desk WW Bangalore.

* Very quite DD area. Some guys talk over the phone even though it's frowned upon. That's an annoyance.

* Conversation is possible even in the quiet zone. I made quite a few acquaintances.

* The hot desk area is noisy as hell, most of the days. There's also WW background music. It just feels like a pub.

* The entire WW area, including toilets is super tidy and I mean tidy like a hospital OT.

Overall pro WW.

Tim Roughgarden is a fantastic teacher. I personally love his style and speed. He throws in some humor here and there and makes learning a lot of fun. His lectures on graphs are absolutely brilliant.

This same course is available on Coursera as well.

Some of the recent developments in Bangalore are heading in this direction. Mixed use is absolutely one of the better approaches to provide a good lifestyle to people. The sad drawback is that you get isolated and will lose the belongingness to the city as a whole. This is not a big deal in the short term, but in the long term, a set of isolated societies isn't that harmonious for the society.

By isolated I don't mean walled, but a natural preference to just be within your conclave.

What might rather work is super fast door to door transit, which in the present is quite unrealistic. Congestion creeps in pretty fast.

While I don't have any direct benefits to enlist, I'm sure there are multiple indirect benefits. I come here, when I am lost without knowing what to do. I'm one of those who reads the comments first and then the articles. So HN acts as a 2 layer curation filter. I came across Andrew Ng's course here on HN and that was an eye-opener, changed my career trajectory. I've learned a lot about parenting, entrepreneurship and startups. I don't live in Silicon Valley, so for an outsider, HN is a good window into SV as well.

Reminds me of this article, about how a single reward could potentially destroy humanity. It's just a thought experiment though.

https://www.salon.com/2014/08/17/our_weird_robot_apocalypse_...

but the DRL framework encompasses all the standard classification/regression tasks and also includes decision making

Except labelling data, correct ? which is non trivial effort.

I think the point he makes is that RL doesn't work as the world believes it to. RL is seen as the holy grail for AGI. He is saying that isn't the case, at least as of yet since RL algorithms don't generalize. For all other tasks, where RL can be used, other existing alternatives fare much better.

Many times, markets and business dictate time pressure. So yes, this is very real in startups at least. Like for instance, an inevitable market development might want you to ship code by the next week, however improbable or impossible. I know I'll get a lot of flak for saying this, but that's ok. I've experienced it first hand. In an early stage startup, when faced with the choice of losing revenue/delivering on time, you'll choose the latter. That means your shipping slot is fixed and so are your features. You'll just have to slog it out or work real smart. Even if you cut features, you'll still realize there's a lot to be accomplished in a short span of time. This is mostly a one off thing. If such events repeat endlessly, that is classic bad management. So you'll need to deliver on time, sometimes :).

So, I guess that companies which administer such time based tests foresee at least one such situation for that role. I'm just speculating.

I can vouch for this. A simple 2 hour script turned into a 2 day affair since production quality code was expected. Thankfully the assignment was spread over two weeks.