tworats

Two had small youtube channels delving into details of ML papers, one had a blog with a good explanation of a specific detail on a paper I was interested, one was a recommendation by my graduate advisor (I keep in touch with the university I graduated from a thousand years ago). The undergrad had a fantastic comment on a thread somewhere, I think in Kaggle.

I've had great luck with graduate students (and sometimes even ungrads): I'll find a grad student who knows the topic, often by searching for lesser known content creators on youtube/blogs, and reach out with an offer to pay for their time to explain a particular topic. So far it's worked out great, and I've created relationships with smart up and coming people in the field - they make meaningful money, I accelerate my learning. I ended up eventually hiring one, and another did consulting for me for some time. We had a routine where I'd ask them to read and summarize / teach me ML papers that were of interest to me, which they in turn could use in their studies/thesis/youtube channels.

Tips on this: content creators tend to be more open to and better at explaining things, and you get to see their ability to explain before you pay them. If you can, overpay them - students need the money more than you do :-)

The Illustrated Transfomer ( https://jalammar.github.io/illustrated-transformer/ ) and Visualizing attention ( https://towardsdatascience.com/deconstructing-bert-part-2-vi... ), are both really good resources. For a more ELI5 approach this non-technical explainer ( https://www.parand.com/a-non-technical-explanation-of-chatgp... ) covers it at a high level.

Coupa | San Diego / San Mateo / Bay area / Remote | Software Engineer, Data Platform Team | On-site with WFH days | Full Time

Join us in Coupa's (NASDAQ:COUP) data platform team to build and extend the core technologies used in analyzing customer data and providing insights. Spark and python for the backend, Django and Rails for apps and services, machine learning of all sorts (simple classifiers to deep learning). If you're a great engineer with experience or interest in data and machine learning and a good understanding of stats this would be a good match.

Job description and application:

Please reach out to me (Tony): td at coupa dot com.

> The rewards for working on new ideas are weighted by the value of the outcome. So it's worth working on something that only has a 10% chance of succeeding if it would make things more than 10x better

This is the primary point here - the "crazy idea" may (and likely does) have a low probability of being correct and succeeding, but with an outsized reward it may still be worth pursuing. This is, almost by definition, the correct way of looking at it.

Also note that he's talking about crazy ideas from "reasonable domain experts", not your run of the mill crazy.

This is not just a response to Mighty - he's been talking about this for years.

Coupa | San Diego or San Mateo / Bay area | Software Engineer, Data Platform Team | On-site with WFH days | Full Time

Join us in Coupa's (NASDAQ:COUP) data platform team to build and extend the core technologies used in analyzing customer data and providing insights. Spark and python for the backend, Django and Rails for apps and services, machine learning of all sorts (simple classifiers to deep learning). If you're a great engineer with experience or interest in data and machine learning and a good understanding of stats this would be a good match.

Job description and application:

https://jobs.lever.co/coupa/5cdd0957-37cd-4419-8537-de60b3bd...

Or reach out directly to me (Tony): td at coupa

Coupa | San Diego, CA | Software Engineer, Data Platform Team | On-site with WFH days | Full Time

Join us in Coupa's (NASDAQ:COUP) data platform team to build and extend the core technologies used in analyzing customer data and providing insights. Spark and python for the backend, Django and Rails for apps and services, machine learning of all sorts (simple classifiers to deep learning). If you're a great engineer with experience or interest in data and machine learning and a good understanding of stats this would be a good match.

Job description and application:

https://jobs.lever.co/coupa/5cdd0957-37cd-4419-8537-de60b3bd...

Or reach out directly to me (Tony): td at coupa

Do these come with stock android, and are they updatable? Or, can you install some other distro on it (eg. cyanogenmod)?

Great stuff. I know very little about music, but found it very helpful to visually decipher and decompose the sounds.

Would it be possible to isolate each instrument? Turn each one on/off?

On the Python side there's Lamson and Google App Engine also has nice email handling capabilities allowing you to parse and access from your own code / app.

http://lamsonproject.org/ https://developers.google.com/appengine/docs/python/mail/

I had the exact opposite reaction - allowing me to play songs for free let me experience the service, so by the time they asked me to sign up I was happy to.

Great to see real world use cases, and very encouraging to see PyPy performing so well. I'll definitely be trying it on my future compute-intensive projects.

For what it's worth I'm a brown man with the occasional beard and I've never felt singled out or degraded when flying. I've been selected for pat downs and further screening surprisingly infrequently, and the TSA have been nothing but professional. I miss the pre-TSA days as well, but I'm no more hassled by it than the average non-brown guy.

Unfortunately getting a "job" will be difficult with a prison record, but consulting could be a good path. If you can find someone to sub-contract jobs to him he can get started (he might have to take a low rate to start with), and once he establishes himself he'll be able to win new deals himself.

As he establishes himself it'll probably be worthwhile to create a company that he can build a reputation around - that way his personal background becomes less visible.

Same: producing when standing, consuming when sitting. I'm actually more productive, and if feels great. I'm also much more mobile when standing - I'll walk around if I'm thinking.

I am more physically tired at the end of the day, but in a good way.

Let me add to his anecdotal evidence - of the angels I personally know (~15), the vast majority are price sensitive. If I expand that to angels I know slightly, the percentage of price sensitivity increases.

You can argue that they're irrational/stupid, or that they're not the investors you want (some of them actually are very good, you'd certainly recognize them by name), but it doesn't change the fact that they exist.

I can't comment on relative usefulness of valuation sensitive and valuation insensitive investors.

Opened a few tabs with HN items in it. Booked my trip to SF, then started browsing the tabs. Hit this tab, tried it, found a flight for about $80 less than I just booked.

Lesson learned: always read HN before doing actual work.

I'm looking for a middle ground - SQL/structured data extensible with non-structured data. The ability to extend sql tables with unspecified json-like data. The love child of mysql/postgres with Mongodb.

Any such thing exist?

A few months ago I was about to get rid of my quite old Galaxy Vibrant, so with nothing to lose I installed CyanogenMod on it. It's worked so well I'm keeping it as my main phone for now - it's clean, stable, and looks much nicer than the Samsung ROM. The only downside is significantly reduced battery life, but I can live with that for now.

Will definitely be trying CM10 if it's supported for the Vibrant.

I've been impressed by Andreedsen Horowitz's approach of recognizing winners in large markets and putting large amounts of money to work. It's gutsy and smart.

However, I'm concerned about putting large amounts of money into startups that have yet to become businesses. An over-abundance of resources is often detrimental to realizing a vision and finding a market.

Anyway, I remain a fan of A16Z and suspect they've seen something compelling not yet visibile to the outside observer.

Really depends on the VC - some of them are very courteous, others downright psychopathic - but radio silence is not unusual. As much practice as most VCs have had saying no it can still be hard, particularly after lengthy discussions, so sometimes they don't say anything.

Love the codecasts (the iPhone one was very informative), but each time I go to cssdeck I struggle to find the play button. This time it was hidden under one of the code panels and I only found it after accidentally dismissing the code panel.

Making the play button more prominent and placing it in the outer chrome would make life for the new viewer easier.

I went to public school in Iran until the age of 12, private school in the UK from 12 to 15, and spent my final year in public school in the US.

Observations:

Math: what I was learning as a junior or senior in the UK/US was what I'd already covered as an 11-12 year old in Iran. The only new subject we covered was calculus. That really sucks and is a big concern for me since my kids go to public school in the US.

Science: was very hands-on in the UK (lots of experiments and labs). Much better for learning than the text book approach in Iran. I only had 1 year in the US, but it was much less hands on as well.

I think you're making my argument for me: if anyone has stats on the total number of adsense ads served vs adwords ads served, we can derive the value of adsense vs adwords ads.

There are many more adsense ads displayed than adwords - it's google searches vs. most of the rest of the web. Yet it only makes up 1/4 of google's revenue.

Last I checked there was an order of magnitude difference between the value of an adwords vs. adsense ad.

I hope he's not actually putting his retirement money on the line here.

It's not about the quality of the ads, which he argues will get better with more data. It's about intent.

When I search on Google, my natural next action is to click a link. That link can be organic or an ad. If it's an ad Google stands to make money.

When I browse Facebook, my natural next action is not to click a link to an outside service/store. It's to continue to read status updates and browse around. Clicking an outside link is unnatural and unpleasant in this context.

This is the same reason ads on other Google properties such as gmail do so poorly - they may know exactly what you're likely to buy and could therefore present you with the perfect ad, but the activity you're engaged in does not naturally lead to clicking on ads.

I started with an early version of Titanium, made a lot of progress very quickly, but ended up hitting many bugs and performance and memory issues.

Each new version of Titanium made steady progress, but didn't really make for a crisp, responsive app.

So I gave up, huddled down with a friend who knows PhoneGap, and rebuilt the app in PhoneGap.

Well, it turns out there are just as many stumbling blocks with PhoneGap as there are with Titanium - if you want a responsive, good looking app, you have to jump through many hoops. I was trading off one set of headaches for another.

I'm now back on Titanium. I've reworked the app significantly and am finally getting to a happy place - it's responsive on both iOS and Android, and looks good. Still a lot of memory usage, but no crashes, so perhaps it's not an issue.

My take at the moment: PhoneGap gets you more platforms and has a lower learning curve. However, it's hard (but possible) to make a native looking/feeling app. Titanium is a pain and requires a lot of discipline in your Javascript coding style, but you can get a nice native looking/feeling app if you're willing to put in the time.

I haven't directly experimented with JQM in a WebView, but based on my experience with JQM on PhoneGap and Titanium in general I would not recommend that path. JQM is slow enough on its own, and Titanium has its own performance issues, so the two together would likely be a challenge.

If you're going for Titanium you might as well use native widgets instead of JQM - otherwise PhoneGap is probably a better choice.

This is conflating so many unrelated things. Leaving work at 5:01 is great - leaving work whenever you need to is great. But spending your life working at "just a job" is a waste. If you're only doing your job to make money you're doing yourself a disservice.

Putting aside the morality of this particular case, I'd point out that this type of copying will discourage early sharing of concepts / projects on HN, which is not a good thing.