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

95%-ile isn't that good (2020)

danluu.com

Post reply on HN
Print view View original

121–130 of 161 posts

Re: 95%-ile isn't that good (2020)

#121

Post by Clubber »

lelanthran wrote: post #89
PragmaticPulp wrote: post #30

Earlier quoted context omitted.

> For a lot of people their career is just a way to get paid. As long as they don't get fired that's enough for them. This was one of my hardest earned lessons when I became a manager. Previously I had worked my way into high-performing teams where everybody really cared about doing a good job. I was so surrounded by people with that mindset that I assumed that’s just how everybody in tech operated. Only later did I…

> If you’ve been surrounded by high achievers and motivated people who would have chosen programming jobs because that’s what they like to do every day, it can be a shock to work with people who really just hate doing anything that resembles work. That's a false dichotomy[1] [1] If I am reading you correctly, you're saying that there's two extremes 1. Programmers who enjoy it. 2. Programmers who hate doing anything t…

>aren't going to be the least bit excited about mindless CRUD duties on an over-engineered stack that has 27 levels of indirection before it even hits the database record.

Yes, this is really the issue isn't it? Most problems have been solved and made into a library. The public internet is coming on its 30th year. Mobile apps are coming to their 15th year. Nothing really on the horizon. Most of what's left is the same drudge work over and over again. Drudge work is fun the first time you do it, but gets boring fast. It's like any artist forced to paint walls for a living. You'll have to excuse them if they aren't overly excited about their 30th year of wall painting.

So, how do you improve?

1. Try to make your wall painting more interesting. Carve out interesting stuff from time to time, even if it's a POC that will probably never get released. This is where you have to be creative/good as a manager and not just a whip cracker. What are different people interested in? Some like automation engines, some like ML, some like experimenting with new frameworks and languages. Experimentation like this used to be pretty common.

2. Give people some real vacation. Developers tend to be on call 24/7. Take it easy with the whip every once in a while. A good month would really relieve the burnout, but to most US companies/managers, this is akin to poison.

Both of us know this probably isn't going to happen, so be comfortable in the bed we've made for ourselves.

Re: 95%-ile isn't that good (2020)

#123

Post by whoomp12342 »

rr808 wrote: post #14

Trivial to be top 5 percent of the general population at programming Possible to be top 5 percent of the programmers Very difficult to be in the top 5 percent of FANG programmers Nearly impossible to be in the top 5 percent of Google Principal engineers

yeah but how do you event quantify such a thing?

Re: 95%-ile isn't that good (2020)

#124

Post by bo1024 »

flerchin wrote: post #110

The observations about 95% percentile being good or not are distracting from the point that focused, reviewed practice is necessary to level up.

True, and the point that people are often weirdly bad at deliberately improving, or weirdly unmotivated to practice deliberate improvement.

Re: 95%-ile isn't that good (2020)

#125

Post by rapjr9 »

In machine learning 95% accuracy is not very useful. It means the algorithm will be wrong 1 out of 20 times. So every 20th word in speech recognition will be incorrect, your medical diagnosis will be wrong for every 20th person, your face will not unlock your phone every 20th time, every 20th person will not be allowed to log into a web site. There are some machine learning applications where 95% is good enough or even great such as web search engines, but for many important applications it is nowhere near good enough and you need something more like 99.9% or better to be actually useful. Think about how often the algorithm can be allowed to make a mistake. If it's never, then don't use machine learning. If it's 1 in a million you need 99.9999% accuracy. 95% looks good in machine learning papers in computer science, but in the real world 95% often means dismal failure. Think of an autonomous killer robot, is mistakenly killing 1 out of 20 people ok? For a self driving car is harming the driver or car (or pedestrian/dog/child/other car) mistakenly even 1 out of 1000 (99.9%) times ok? I think not. One measure of what accuracy you need is how well people do the same task. If you can do better than people most of the time, maybe the algorithm is useful, though you also have to compare exactly how the algorithm fails to the way people fail. If people fail by choosing a fender bender and the algorithm fails by killing someone, the algorithm has to work better than people do.

Re: 95%-ile isn't that good (2020)

#126

Post by throwaway6734 »

the__alchemist wrote: post #117
throwaway6734 wrote: post #113

Earlier quoted context omitted.

What's your current elo?

1150 in Aoe2, 1250 in Aoe4.

Oh cool, I'm at similar level in AoE2.

I've found that if you consistently make vills & army you can compete pretty easily at this level. For the most part I play random civ, don't use hotkeys, and don't use build orders and it hasn't been too rough.

Lately I've been playing more teamgames and spending more time watching streamers play (shout out to survivalist!).

Re: 95%-ile isn't that good (2020)

#127

Post by jvanderbot »

rapjr9 wrote: post #125

In machine learning 95% accuracy is not very useful. It means the algorithm will be wrong 1 out of 20 times. So every 20th word in speech recognition will be incorrect, your medical diagnosis will be wrong for every 20th person, your face will not unlock your phone every 20th time, every 20th person will not be allowed to log into a web site. There are some machine learning applications where 95% is good enough or ev…

95% accuracy has nothing to do with being in the 95pctile of active participants other than using the number 95.

To achieve a 95th pctile ranking among the relevant machine learning frameworks means being more accurate than 95 % of other approaches, and that's probably quite good and useful.

Re: 95%-ile isn't that good (2020)

#129

Post by jvanderbot »

xigency wrote: post #17

Admittedly, I only skimmed this. But within the space of video games, I’m not sure Overwatch is a stunning example since it’s generally a casual game. I’ve been playing various iterations of Counter-Strike and StarCraft for decades. Sometimes with deliberate practice, including reading strategies, playing custom maps to practice certain skills, and watching series aimed at improving level of play. And despite the hun…

TFA: The "one weird trick" is that, for a lot of activities, being something like 10%-ile among people who practice can make you something like 90%-ile or 99%-ile among people who participate

An important point is that you're now better than 95% of starcraft players if you reach middling competitiveness against those who compete regularly. At 50th pctile in competitions youre absolutely better than the vast majority of starcraft players.

Re: 95%-ile isn't that good (2020)

#130

Post by jjoonathan »

coldtea wrote: post #77
xboxnolifes wrote: post #70

Earlier quoted context omitted.

In every competitive game I've ever played (Overwatch, CSGO, LoL, TFT, Rocket League), the top 5% definitely wasn't "that good". In my mind, the top 20% are people who actually play. If you are stuck anywhere below that, you aren't even trying to improve. If a leaderboard says you haven't even cracked top 50%, then that leaderboard is definitely cutting players that don't play much from the bottom of that list. In mo…

> Then the top 5% is when you start understanding the game at a competitive level. The very start. On the other hand, your handle is "xbox, no life". The whole OP premise is: "Reaching 95%-ile isn't very impressive because it's not that hard to do. I think this is one of my most ridiculable ideas. It doesn't help that, when stated nakedly, that sounds elitist. But I think it's just the opposite: most people can becom…

Ah, the iron laws of gaming strike again!

* Anyone who is worse than me is a scrub

* Anyone who is better than me has no life

Post reply on HN