quantumofalpha · HackerTrans

quantumofalpha

no profile record

コメント

5 年前議論

Typical red LEDs used for displays have a peak emission somewhere in 600-650nm range, but it's bell curvish so yeah you can expect a small amount of 670nm light out of it.

Does getting exact 670nm light matter? It's just the wavelength that this particular study investigated. They do not offer conclusions for other wavelengths as they did not test them, it's as simple as that. But a lot of other researchers did, especially in red/infrared part of the spectrum and generally with positive results - http://bitly.com/PBM-database. So I would guess you should be able to get some of the described effect at nearby wavelengths as well. 670nm seems to be near an absorption peak of cytochrome c oxidase, a molecule believed to be involved in the mechanism of light therapy action, which is why I suppose it appears frequently in research.

5 年前議論

Have a look at http://bitly.com/PBM-database - if there's one it's probably listed there

5 年前議論

It's sign that your VPN works:) Unfortunately that's the price to pay for anonymity today. Go to another website? DDG instead of google.

5 年前議論

GPUs are too slow for HFT trading, yes. Deep neural networks in general do work in this domain, people do use them, but the stuff that you can profitably deploy into production is not going to be your vanilla garden variety neural networks, there's a lot of extra engineering required to make it fast enough.

5 年前議論

Really depends on time horizon you're talking about. They actually work pretty well for HFT, there's plenty of data around, and most of the information is just in market data - no nasty low frequency stuff to deal with like news, earnings, alternative data, insider trading, butterflies flapping wings in china etc. But the problem is by the time your GPU spits out a datapoint somebody else can go in and trade a few thousand times in the meantime. State of the art on the most heavy competed exchanges is that your fpga (or even asic) with a fiber connected directly to the exchange needs to start sending ethernet/ip headers even before it made up its mind what it wants to send in the payload.

At lower frequencies when the data gets thin and noise/overfitting is a major problem, yeah it makes sense to use simpler models. Bias/variance tradeoff in action.

5 年前議論

Why not both? Linear models for absolute lowest tick-to-trade latency doesn't preclude you from using fancier stuff at earlier modelling steps. Final linear models you ship to fpga can be mere distillations/triggers

5 年前議論

Amazon is worst payer among them in the US as well. Facebook and Google pay decent money in London and Zurich, almost on par with US in the latter, London is maybe ~30% less.

> How much are you left with now, ~1% of the total tech jobs?

If you want top 1% pay, you gonna have to hunt for top 1% jobs, seems obvious, no?

5 年前議論

The author is spot on. You just happened to be working for the scrappiest/shittiest faang that's all. Shop around, there is decent money to be made in Europe - but of course it's not as easy to walk into it as in the US today. Oh and don't forget that it's TC that matters in big public tech, not salary.

5 年前議論

> The main issue is that there is just nowhere near enough housing to keep up with population increases, seemingly anywhere in the Western world.

With falling fertility rates, natural population growth in the west is about zero today. If not even (going to be) negative.

5 年前議論

Exactly. If you're in tech, you should be shooting first of all for $500k TC today than $500k pnl, the former is way easier (luck aspect aside). No point in thinking about active investing until you have like 7 figures of own capital to manage.

5 年前議論

Put spare money every now and then into VOO or VT or similar and forget. No knowledge needed and time investment is like a few minutes of your time a month/quarter/year. Squeezing more alpha than that and not losing it, properly done (i.e. not gambling), is at least a part-time job. Chances are you'll do better putting that time into your own career.

5 年前議論

wat? Google used clicks ("navboost") since approximately forever, it's one of strongest signals for all major web search engines. I guess the better question is why they don't optimize ranking directly on clicks, why still bother with all that expensive human raters business... well optimizing for clicks helps but only until a certain point beyond which it starts hurting relevancy by overpromoting old popular results and clickbait. But as an ingredient in overall ranking in one way or another they and everyone else definitely use click data.

5 年前議論

Steadily losing 0.5-1kg/month just by cutting out carbs and sugar. Not really full keto (which requires like 70-80%+ calories from fat to maintain ketosis), just paying attention to nutrition facts, stopped buying high carb food, started cooking more with quality ingredients I control, otherwise eat whatever and however much I want.

The biggest change by far was cutting out cow milk which is actually very caloric and sugary. I easily drink 1l+ carton a day and that's 500+ cals alone! Human metabolism is pretty high but that's just too much calories to consume every day and that you can hardly feel. Found a good substitute - almond milk, especially Blue Diamond and Alpro brands - they cream coffee and taste in it just like regular milk with only a fraction of calories and no carbs. Another good sub I grew to love is grated/mashed cauliflower instead of rice and potatoes.

5 年前議論

Not flaky at all. 10M+ LOC monorepo with a few hundred devs here. All it takes is a fast build system. Full clean rebuild for me is a few minutes tops (distributed on a shared cluster), incremental is O(seconds) - mostly linker. Precommit checks and code review system both warn if a build is failing, so there's no reason ever but sloppiness to check in broken code. Full test suite on other hand takes hours to finish and so gets broken half the time - the longer the feedback loop the more likely the thing is broken.

5 年前議論

> meaning I would need sponsorship for an H-1b work visa

There's your problem. H1B is just such a difficult and lengthy process today. Even for experienced candidates - I have a decade yoe at faangs, still couldn't crack it recently. And in your position, without experience and a degree, yes you're just wasting time.

You have to apply for H1B in March for a sub-30% chance (it's a heavily oversubscribed lottery) that you might get to start in October. Few companies are willing to go through it to import brand new overseas talent. Most just sponsor people already having permission to work somehow (H1B transfer from another co, F1 OPT, L1, through spouse etc) to retain them in the states.

Some alternatives of what you could do:

  * Go study in europe on the cheap. An undergrad degree is not a must for programmers, but still a very important piece of paper that you need to open doors later in your career. It's actually even a requirement for H1B: degree or 12+ yoe.
  * Go for masters afterwards in the US, not cheap but you can stay work for 3 years on STEM OPT and no lottery involved as with H1B.
  * Work for at least a year for an american company in europe, then you can transfer to the states on L1, lottery free.
  * Go to Canada, they can issue you a passport in 3 years with which you can come work in the states on TN visa, also lottery free.
  * Find second half with the right type of passport / visa that allows you to work in the states...

5 年前議論

> So one option is to move to one of the outsourcing companies or even better to a smaller company dealing with western clients.

That's pretty much what most (if not almost all) of ukrainian tech is doing. And coupled with low CoL and widely abused tax loopholes, they are making bank, netting comparable money to 6 figure jobs in the west. So I'm just saying that eastern europe == poorly paid is a misconception.

5 年前議論

You are grossly underpaid, even for eastern europe. Guys just across the border (or two) from you make 2x+ more, while paying next to nothing in taxes (5% I've heard, as faux contractors) and enjoying some of lowest CoL in europe - https://jobs.dou.ua/salaries/dynamics/

5 年前議論

Money can buy you financial freedom, more holidays than you'd know what to do with, good place to live almost anywhere in the world, trust fund for your heirs and all that other stuff you mention. Good money - the kind you can reasonably fast and predictably accumulate in California today, say ~decade at FAANG, but have near 0% shot at reaching in Europe. To me being able to retire in 30s and have a lifetime to yourself, own pursuits and family is much more enticing than a lifetime of neverending 9 to 5 grind no matter how chill it is and how much parental leave they give (it can't beat ∞). But indeed generally Europe is a better place for the less successful/lucky. Not everyone is ambitious and lucky enough to make it in the US, especially outside tech. Good thing capital is mobile, you/your kids don't have to stay in the US forever once you've reached your target.

5 年前議論

Still pretty useful in high noise/signal regime. Though at my last two gigs in this area they were kind of obsoleted in favor of generalized additive models which are more interpretable, robust and have more powerful fitters e.g. modern boosted (xgboost/lightgbm/whatever), bagged, depth 1-2 tree ensembles then basically fit curves to scatterplots.

More generally in industry the bottleneck is usually data, not models.

5 年前議論

Kind of, I've seen some non-public Intel SKUs in the fleet but afaik they just had tweaked core counts and clocks for better perf/watt. They were also exploring non-intel architectures at some point, ppc especially but no significant prod deployment yet. And then I'm sure you've heard of special purpose hardware like TPUs