quants-n-traders

[hardy marsh]

have u gone thru it

[dire arrow]

i havent done shit ngl

need to start asap

if i flop da js phone im jumproping

[hardy marsh]

lmafoo same, im so late, i havent taken prob or combo yet either

[hardy marsh]

ur getting it manifest that shit

did u do sig disc day

[dire arrow]

ion really get shit for disc days since im not t10

i j got lucky with js

[hardy marsh]

js phone is so fye bro

praying for u

[hardy marsh]

gotcha

[dire arrow]

that shits in like july tho

if i fail its fs on me

hopefully I pass the optiver future focus tech

[hardy marsh]

wowww u got time

[hardy marsh]

i missed the boat on that so tough

hoping for sig atp

[dire arrow]

u got it twin

[hardy marsh]

what would be more important for me to take next sem? combinatorics or prob? I'll eventually take both but as a current soph trying to target a qr internship for next summer what would be more important to learn earlier?

[jaunty root]

lemme know if you down to clean some rag. 💸

[warm sinew]

guys i was wonering if CS + STAT is better or Mathematical Finance + STAT is better for breaking into quant trader

for bachalor's degree

[lethal apex]

Doesn’t matter much tbh, but if you don’t have a CS background it would be useful to take a few classes

Skills necessary for trading have so little overlap with any major it’s actually pretty funny

[warm sinew]

i understand now, thank you thank you

Do you have any advice on what types of co-op I should be looking for during my undergrad before doing a quant trading co-op
iam in second-year now, and i've done one machine learning coop last coop term

[lethal apex]

quant soph program > risk/trading at bank > Faang second year

Prioritize in that order

[thin quest]

aren't the soph program's usually discovery days

other than cit launch

[lethal apex]

yea and those are top priority

Also there are definitely more and small quant firms may have internships

[dire arrow]

all the bank roles so poor tho

[hardy marsh]

thank you!

I wish, just can't fit both in the sched at the moment. I will prob have to self study over this next sem, then take the class I couldn't next fall.

out of curiosity, was wondering what u guys think -- which firms have the best culture (good WLB, dont hire and fire quick).

[lethal apex]

You not going for tc as an intern just prestige

Unless you in the lower tax bracket and trying to feed your family

In which case you probably won’t make it in quant

[dire arrow]

how corny is it to put discovery days/insight weeks in projects/other exp section on res

or is it p normal

[dire arrow]

I disagree

plenty of quant ppl r j random midwestern/immigrant kids with 3sig math ability

[thin quest]

What career isn’t working class then

[dire arrow]

starting a pe family office

[lethal apex]

who lied to you

You know there aren’t just dirt poor and ultra wealthy right

??? Bro said average and 80-95 percentile nw

in the same sentence

most people grew up middle/upper middle class, more recently upper middle

[dire arrow]

while yall here throw me a lil thumbs up/down

[lethal apex]

Read what I said here, referring to the bottom tax brackets, people who have to work to support family

[kindred mortar]

Is this due to quant just being massive stem nerds which is mainly due to genetics

[thin quest]

Is owning real estate working class?

[kindred mortar]

Plus huge immigrants / second generation immigrants

[thin quest]

You still have to monitor your tenents and such

[dire arrow]

is stem iq truly genetic

[thin quest]

Unless you outsource mangement

[dire arrow]

I improved my working memory a lot j with practice

but then again my parents are quite smart

[lethal apex]

You must struggle at logic puzzles. “You will likely not succeed if you are poor” != you need to be ultra wealthy to succeed

[thin quest]

So I’ve been doomed since i was born

[kindred mortar]

Just buy stocks

[thin quest]

Is a CEO working class?

Their income is directly correlated to the quality of their labor

Their equity compensation is directly correlated to certain company targets

a better ceo would get better pay

We also see 0.1x engineers get 400k a year

Labor is more than technical work

Leadership is also a kind of labor

Value creation is based on quality of labor

I agree

Let’s postpone this convo

Gotta grind greenbook

[dire arrow]

yo twin

what chap u on

does anyone here know if its aight to skip the stoch calc part for qt ints?

id be really surprised if they asked that to undergrads unless I indicated somewhere on res that ive used it before

[dire finch]

this guy really pulled out the dictionary and went "wait everyone is working actually 🤓 "

[cloud root]

yea

for sure

like hes not wrong

wait

no thats stupid asf

braindead ass take

i read it as ppl in quant wont make gen wealth

not that ppl w gen wealth dont want to do quant

thats idiotic

ppl w gen wealth already hve the money

[daring basalt]

is there a newer doc with resources someone has besides whats pinned here

[cloud root]

so they want nothing more than prestige

and quant is like hte higest prestige in tech/finance

[cloud root]

for what swe?

yea

[daring basalt]

developer yeah

[cloud root]

https://leetcode.com/

there u go

[daring basalt]

bruv

[dire finch]

ooh a disagreement is unfolding

[dire arrow]

all the real deep pockets mfs I know go into pe or vc

[dire finch]

agree

if you had generational wealth you'd likely go for a career with high publicity with disregard to income

if you had the talent to do so

oh and provided you're not aut like me

kek

make that money work

this guy dogwhistles

[cloud root]

obs not

but the prestige of being able to work at renn and 5rings or 2sig

is def higher than working at a big 4 consulting

or like at a bank

and also thats the prestige

that is flashy

working at a family firm isnt flashy

and thats what these ppl want

im well aware

im surounded by those ppl

ik what they want

im aware gang

ok let me put in this in perpective for you ok

one of my investors chartered me a jet and flew me and my partner out to his office

and withing one meeting

he gave us 5m

were 2 kids mind you

this guy shits money

his kid wants. to work in quant

its not the same

was it your own firm

or was it a big name

no

that actually happened lol

like i get what youre saying but u also forget that theres different kinds of gen wealth

a good ammount of kids born into that want to prove to their peers/themselves

that theyre more than just daddies money

[dire finch]

pffft you should visit

[cloud root]

and for that they need to get flashy jobs

and theres nothing more flashy in terms of prestige than renn tech

[dire finch]

surely that's a banded phenomenon

[cloud root]

at first i said a majority but i realizeed i was wrong

[dire finch]

like only ppl within [x, y] care about that

[cloud root]

so i corrected my self to saying a good ammount

[cloud root]

yea but thats the ppl they want to prove themselves to

they dont care about the opnion of random discord user 1235

[dire finch]

meh i underestimate human psychology

[cloud root]

they care about what their fellow classmates from their swiss boarding school think

[cloud root]

me too im a psych major matter fact

[dire finch]

i can't tell if this other guy riffing fr or not

[dire finch]

yeah fair enough

but there's plenty of prestigious careers that don't require high iq

[cloud root]

yea

for sure

but the original argument was that ppl w gen wealth dont wanna be qrs or qds and shit

and i was just stating why i dont think thats true

obs we cant generalize a whole group of ppl

i was just saying that there is a group of ppl who exist who are gen wealth and want to go into quant

and its a sizeable % of the area it self

[dire finch]

they're probably less likely to go into quant cuz gen wealth people (outside of academia) don't care about academics nearly as much beyond the "please pass boarding school and get into ivy oxbridge" part

[cloud root]

but nuts for brains overhere cant understand someone elses opnions without getting mad and just breaking down

and just assuming everything i say is a lie just for the sake of his argument

[dire finch]

i agree with this, idk how significant of a change it is from overall pop of quant eligible people

[cloud root]

again i cant generalize a whole group of ppl but theres a good ammount of ppl who do care about academia not just bc of academia itselgf

but bc they want to prove themselves

even if its subconsciously

[dire finch]

right right

i know a guy like that actually

[cloud root]

me too

i know a couple

[dire finch]

his family is 700 years descended from some territory in germany

and owns multiple castles

[cloud root]

oh nvm

not the same level of guy ik lol

[dire finch]

but hes studying engineering to be quirky

[cloud root]

the one ik is just like his parents were start up ppl who got a couple hundred mil

and now hes trying to break into quant(he has in fact done it)

[dire finch]

oh for new money people it's probably tougher

[cloud root]

yea thats just the example of i a friend i personally am close w yk

but like it applies

[dire finch]

gotta prove they're not a dud

[cloud root]

yea

[midnight scroll]

would you buy one

If I had mad cash i'd lowkey cop a loro piana sweater or something

if it pills im kms though

mfw I drop 10 bands and it pills

[thin quest]

End of 4

[cloud root]

i thought if u washed/dried it how they instructed

it wouldnt pill

[neon mirage]

Hi quants and traders

Does anyone have any good resources on learning cpp outside of watching cppcon or reading the STL

Thanks rift

[lethal apex]

Yes… you are missing quite a bit by being poor.

[brisk ingot]

tour of c++ is pretty good

[lethal apex]

You don’t actually believe quant is a meritocratic right? The industry is one of the better systems humans have developed for finding merit, but it’s nowhere close to perfect.

[brisk ingot]

but otherwise just do c++ projects

[lethal apex]

Prep is not free

[brisk ingot]

and try to use as many stl/modern features as u can

[lethal apex]

What does blackstone have to do with anything? We are talking about quant

[neon mirage]

Thanks void

[midnight scroll]

yes its quite expensive 😂

they have some shit products tho

They started out as a fabrics manufacturing company

its all marked up a lot obviously and whatnot

But they use some specialty fabrics

[lethal apex]

And you wonder why I questioned your logic abilities. One industry being insular doesn’t exclude another from also being insular. There are degrees to this and my claim is that if you are not in the top 10% of wealth in the US, succeeding in quant is virtually impossible.

[midnight scroll]

LVMH owns them now

wealthy

[lethal apex]

Need cashmere wool sweater desperately

[midnight scroll]

the pricepoint is beyond larp 😭

[lethal apex]

Loro piano is just above larp

[midnight scroll]

yeah gucci si like larp

this is you gotta actually have bands lmao

[untold prawn]

their better coats are 20k you can't larp lol

[kindred mortar]

Hedge funds are top finance

[midnight scroll]

thats not larping rich

if you have 10m in crypto

[lethal apex]

? My guy that isn’t larping wealth tho

[midnight scroll]

ur still rich

I think the concept of old money is kinda silly

money is money

[untold prawn]

idk a lot of tech executives wear stuff like loro, zegna, brunello

[kindred mortar]

Unfortunately you are going to be a option market maker tuning parameters lol

[untold prawn]

this can't be serious 💀

[kindred mortar]

Referral?

[lethal apex]

You are trolling lmao

my fault I fell for it

Old money doesn’t exist it’s purely perception.

[kindred mortar]

Tbf probability of ending up in the right pod is zero

So that’s fine

[lethal apex]

Closest thing you get to it is British royalty

[kindred mortar]

Depends where ?

No shit

I guess 90% in NYC is definitely higher

[lethal apex]

Ability to not work doing high school, ability to pay for tutors, ability to invest in top resources, pay for top university.

[untold prawn]

200k vs 50k is like comfortable vs broke

it is literally a different world

[lethal apex]

Yes? That’s absolutely true

[kindred mortar]

That’s tier 1 firm bonus vs tier 3 firm bonus

[lethal apex]

means paycheck to paycheck vs time to study for putnam

[kindred mortar]

That’s like JS vs IMC 😂

[lethal apex]

lemme guess you watched Goodwill Hunting and you think IMO medalists don’t study

200k@500k im ngl he might be trolling

[untold prawn]

400k @ 300k

[kindred mortar]

Pls make it a bit wider

[untold prawn]

nw > 1M not even a question

[lethal apex]

he freudian slips saying average is top 80-90% nw which suggests he lives in a gated community

so maybe less physical assets more index fund type

[kindred mortar]

Work at market maker and invest in their failing strategies

[lethal apex]

You right old parents = 2 mill ngl

[untold prawn]

still yes

[lethal apex]

alright final market 1.8@2.2 mill

Trust fund only a small percentage, I forgot

yea i just assume you aren’t being serious tho

[untold prawn]

trust fund 5M@2M

😂

[lethal apex]

Do you think sports is also meritocratic

[untold prawn]

yes u can pray to god and be built like wemby

wake up like sung jin woo

[lethal apex]

Lmfao

nah fr tho those sports coaches are so expensive

[untold prawn]

the new nba class is like half nepo babies 😂

[lethal apex]

200/hr for a D3 coach not even a solid one

I’ll walk into a firm right now and I guarantee you it’s mit mit mit harvard stanfors mit mit mit

oh look at that

Another mit

Unc hasn’t applied to college in 30 years or sm

look at the applications of people getting into t10 nowadays it’s wild

501c (definitely not parent funded) generating 2 mill revenue

cure for Alzheimer’s

Yep but some prep high schools go all out

they got connections to real charities

now you getting it

Oh like tj esque

yea that shit is crazy

One org had 3 ceos

Another actually hired employees and then fired them which is wild

messing w people’s lives is actually nuts but idk if I’d ever work for a highschooler

atp job market so bad for mid (awful) swes

covid was the golden age for slacking off

[stone vector]

3 CEOs

[lethal apex]

so I could see it

[stone vector]

lololol

make it board of directors

[lethal apex]

Lmao I’m imagining succession type drama at a highschool 501c, they should make a show about that

Super senior finally graduates to gtech

[stone vector]

it's all bullllshit

bullshit

[lethal apex]

leaving his 501c ceo spot to 5 new juniors

who will make it into mit? Find out tmr at 10

Just the Office except it’s the highschooler 501c

they all working after school

[kindred mortar]

Firms literally don’t even go outside t10 or something

[lethal apex]

? Dawg still adding to the old argument lmao

we alr agreed rift is a nepo baby

[kindred mortar]

My discord is so bad, messages are loading in wrong order

Leetcode engineers fucking up

Who are you? Kendal Roy

[lethal apex]

Needed to prevent my siblings from inheriting the family home

[cerulean bloom]

I don’t think quant is unreachable for non wealthy and people attending top schools. In fact I think quant is a field that is somewhat of an equalizer since you can prove yourself outside of connections and school prestige.

However at the end of the day being wealthy is a massive advantage. Partially resources but primarily network and information. The vast majority of people who even know what a quant is are going to be pretty wealthy. Most of the extracurriculars and resume items that would help you break into this field are done by primarily wealthy people. Not that average income people can’t do them, but most people in the middle class don’t even know what math contests are. Everyone I know who is wealthy (and interested in this stuff) enroll their kids in every extracurricular with the top network since middle school. There’s a reason why the greatest correlation of how prestigious a school you attend will be your family’s wealth

[dire finch]

i didn't know you could pay for math camps and shi

[thin quest]

most are paid

free ones are invite only

[dire finch]

that's crazy

i coulda been more orz if i had 10 bands

[fresh swallow]

The in group in trading is people who heard about it as a career at all

[fresh swallow]

you guys live in a bubble

The most common response from telling the average person you work at any quant/hft is "What is that"

They know what banking and PE is tho

[lethal apex]

Yea, that was my whole point. The fact that people can’t realize it is just because they still cling to the myth that quant is meritocratic

these hollywood “poor genius but also never spent time studying” movies have done immense damage

Which also happens because usaco plat/usamo folks seem to enjoy larping about not having studied ever

[fresh swallow]

Living in the sticks on long island, very flashy

[cloud root]

at least read the whole convo gang

i specifically said

that theyre trying to impress the ppl they went to boarding school in swizterland w (just an example)

and not random mfs off the street

i agree that its a bubble and the avg person wont know what it is

i never said they would

i also said that the avg person is not who theyre trying to impress

so yes you are right it is a bubble

but that doesnt take away from my point

[fresh swallow]

why you care so much about this

[cloud root]

i dont

i just love talking to people

and disscussing different point of views

also i was very bored

[fresh swallow]

We aren't people we're little nicknames and pictures on a computer

[cloud root]

ok mr solipsism

[neon mirage]

Very nice of you to throw 2s in there

[cloud root]

it was just the ones that came to mind in that very second lol

no real reasoning

[neon mirage]

🤫don’t ruin the moment

[dire finch]

im sure they're still plenty fun to work for as swe

[rose holly]

fe or be?

[dire finch]

ohey i had similar dilemma

i chose stripe

[sand willow]

choice has been made long since LMAO

[karmic valve]

https://x.com/deedydas/status/1882104649668530586?t=Wqkcl2axncFC5cfyfwGq4Q&s=19

Deedy (@deedydas) on X

The German national who killed and was killed in a shootout with a Border Patrol Agent in Vermont was in the US on an H-1B.

Felix Bauckholt was a Quant at Tower Research from UWaterloo and a former Gold medalist at the premier high school coding contest, the IOI.

Crazy.

X · Jan 22, 2025, 4:34 PM

yo this is crazy 😭

[lethal apex]

He was cracked

Is working at tower that bad

[strange loom]

reminds me of the yale math guy shooter

[pulsar scarab]

if this is what will happen to me I am canceling my tower interview

[strange loom]

was overstaying visa, likelihood of deportation and subsequent ban is high

there was more than one? lol

thought it was envy

[karmic valve]

wonder why his h1b wasnt renewed/he wasnt in gc process if he was top talent

[north meadow]

Anyone know how many rounds Optiver has for trading? I got the invite for a second technical round. Is the final after this?

[calm silo]

Does meta SWE pass a lot of screens?

[dire arrow]

drw oa was rough bro

[thin quest]

Discover drw?

Kinda light

7 q’s 20 minutes

[tranquil sand]

Anyone currently in/done the process for Optiver SWE intern for this summer? Have first technical soon

[dire arrow]

first 5 were p easy

the tetris one stumped me

i think i j guessed 28

hopefully the bar aint that high for sophs 🙏

[stone vector]

rip

[fresh swallow]

I don't think he worked at tower still

Crazy text to receive though that some guy who worked at your company was in a shootout with border patrol

[lucid cloak]

Did he get fired at tower

[cursive atlas]

So many reports seem to not mention anything specific about the girl for some reason

[lucid cloak]

That would explain why his visa expired

[midnight scroll]

shes not dead, I assume her identity is held private

[hardy marsh]

what i didnt even know this was a thing, when did it come out?

[low vault]

govt confirmed that visa was active apparently, so the article is wrong

[fresh swallow]

Idk but his account was already deactivated on slack when I looked and saw this yesterday

Maybe they were just that quick with it

[lucid cloak]

Interesting

[thin quest]

no like after doing the oa

is there a tech

[brisk nova]

i just did the oa like 5 min ago lmao

[dire arrow]

yo delete LMFAO

i got wrong i dont want mfs to do better than me

[brisk nova]

my b my b i forgot

[hardy marsh]

when did u apply + get OA

[brisk nova]

applied 2 weeks ago oa yday

[hardy marsh]

ughh I did not know abt this LMAOO noooo

is there a resource that has all of these programs somwhere? Ik abt IMC, SIG, Optiver, and Jane Street but are there any other programs out there?

[dire arrow]

i wish

the only way I hear abt these things is if someone mentions it in this chat

nobody at my school recruits quant n shit

[hardy marsh]

just gotta wait for launchpad ig

[dire arrow]

launchpad will save millions

[late root]

Anyone here done a sys design question for quant swe for intern? Have upcoming interview and have never done a sys design before, do they tend to ask sys design related to the job? like design market data aggregator, a stock exchange, order management system

[dire arrow]

ion think its expected to know much market specific stuff

but prob good to practice wit dat

[late root]

ok cool thanks

[lethal apex]

For the examples you named, know them inside and out

[late root]

Sounds good

Best case they ask me that tbh

I dont wanna be asked design google docs lol

Thanks for the advice

[tranquil wagon]

Most firms that I’ve seen don’t expect much/any financial knowledge from swes

[late root]

Really from most ppl i talked to, even for swe they need to understand the intracacies of trading systems

At least the software part ig they dont have to fully understand how the strats work

[lethal apex]

He’s yapping, they don’t explicitly require knowledge of finance but knowing basic order book fundmentals (L1, L2) as well as the engineering requirements of firms (redundancy, latency, etc) goes a long way.

[tranquil wagon]

I’ll be interning at a hft firm this summer and I can barely explain how what option is lol

[late root]

Bruh no way

Thats a w tho

[brisk ingot]

it helps but most good firms will have education in the first week

[tranquil wagon]

I have no clue what L1 or L2 mean here, I have no finance domain knowledge whatsoever

[tranquil wagon]

^

[brisk ingot]

abt all that stuff from the ground up

[late root]

its like order matching priority for a book

[brisk ingot]

is it not that l1 is just bbo

[late root]

Oh nvm im tripping its the lvl of detail in data

ya mb

[lethal apex]

Congrats on akuna

seriously tho that’s huge but how did you dodge those god awful system design q

[tranquil wagon]

I just got a lot of in depth c++ questions and general programming system design stuff, nothing about trading

[late root]

wdym like general sys design

like more common sys design qs?

non spec to trading

[lethal apex]

you mean implementation?

like build vec or whatever

[late root]

yeah or like ds implementation

[tranquil wagon]

It was mostly verbal

[late root]

oh shit i have drw final too

[tranquil wagon]

Thanks

[lethal apex]

Might have the commit seppuku then the hedge fund I interviewed w asked me order book shit

[tranquil wagon]

Damn that’s crazy

[late root]

ig its very team dependent

[lethal apex]

nah it was light tbh I watch way too much youtube 😭

I suck at c++ tho ya’ll are cracked

memorizing stl

[tranquil wagon]

Sooo much stl trivia yeah

[lucid cloak]

whats a hard one that u got

[late root]

ngl i think the cherno is good for cpp stuff i like his videos

[lethal apex]

the funniest thing is people going into python interviews not realizing they ask python trivia

Cython type bs

[tranquil wagon]

Nowhere near enough depth but rlly good for getting started yeah

[lethal apex]

any good textbooks?

or just memorize stl

[late root]

bjorn strousroup or whatev his name is

[tranquil wagon]

I really enjoyed Effective C++ and Effective Modern C++ by Scott Meyers, he has a great writing style

[lethal apex]

Been using python for a decade never once needed a generator

[tranquil wagon]

I think best thing to do is just use the language a lot for non trivial tasks

And you’ll naturally run into most important things

[lethal apex]

Exactly mfs be asking about performant python 😂

[tranquil wagon]

Python trivia sounds crazy

[lethal apex]

What do you classify as non trivial, anything specifically?

[tranquil wagon]

I spend a lot of free time working on my game engine, I use a lot of interesting language features for that

[lethal apex]

Agreed until I listened to a podcast where the dude was making 40 microsecond trading systems in Java

It was 2012 so early days of hft

Or no maybe 2006? I forgot

2012 was knight capital crash so maybe not early days

[tranquil wagon]

If you’ve tuned your jvm properly and are using it right you can get performance that’s honestly not far from C++

There’s a great video on YouTube from a guy talking about low latency Java techniques

[lethal apex]

love C i code in it regularly but c++ is a huge jump

[tranquil wagon]

Yeah there’s a lot more complexity in c++ for sure

[lethal apex]

Learning best practices (low latency industry standard) seems like a never ending goal

[tranquil wagon]

I think you can probably get passable at C++ after working on it for 3 months

Depends how fast you learn ig lol

Not many people get “good” during that time in my experience

Like decent

Idk it’s all subjective

Different people have different bars for “good”

3 months is probably a tight time horizon but i guess it’s possible

[lethal apex]

Arkansas ngl you not wrong to be facing imposter syndrome but I’d say relax and just get started now

[tranquil wagon]

You need to work on projs, you can’t get good at writing c++ without writing c++

If you just study it all you’ll be good at is trivia

[thin quest]

what in 3 months

[lethal apex]

Fr or joke

[tranquil wagon]

Oof late team match

[lethal apex]

oh gotchu no teammatch yet

That’s fire tbh you read beej’s guide?

[tranquil wagon]

Thats a great guide

[lethal apex]

its p famous good for C prog and networking

He mogs you, watch your tone

[tranquil wagon]

This is the face of a mogged

[lethal apex]

@mint aurora python trivia

apparently python caches integers from -5 to 256

the longer you spend reading up on python trivia the longer you realize this language was a hackathon project that lasted a decade

-5 to -1 for error codes apparently

wild goose chase but there is 0 chance you would ever know this unless you knew random Cython commit history from 2017: https://hg.python.org/cpython/file/tip/Objects/longobject.c#l5343

saw on csmajors or sm and tried it out in my terminal

@lunar thicket when you become an interviewer make sure you frame it as an easy python interview then pull this one out

[dire arrow]

can yall stop gapping me

gimme a few to catch up

[thin quest]

yeah it's over for me

[lethal apex]

going down a rabbit hole but this is disgusting: https://www.w3schools.com/python/gloss_python_for_else.asp

haven't morphed it into a problem but basically scroll up to the screenshot of a is b

speak for yourself, lmao

need jump and goto in python

turn it into asm

[dire arrow]

do yall fw Go

my next internship primarily in go

Ive heard mixed things

[lethal apex]

Something awful like
def func(): ...

for i in range(N):
for j in range(M):
if i > j: goto func

don't ask me why it's useful we just need it

[lethal apex]

Goated

[vivid trellis]

Anyone got advice for HRT phone?

Algo dev

[hardy marsh]

intern or ft?

[vivid trellis]

Understood

[vivid trellis]

Intern

[low vault]

I've seen this trivia fact being tossed around

its kinda funny

[lethal apex]

Yea it was on the csmajors reddit so the ngmis have discovered it

The caching details for all different types is still a good trivia q

Even a simple q like how python lists are implemented could serve as an easy filter

[low vault]

tbh i think most trivia questions are kinda bad

like its mostly one-off knowledge that doesn't really matter

the only "good" one that I can come up with off the top of my head is the ones about reassigning lists and whatnot

since you need to understand how python handles references/etc.

AFAIK python is like java where objects are pass-by-pointer, so you need to explicitly copy things when you reassign

when I was 14 I was always wondering why I could do

World w = ...;
w.setBlock(...);

and that would actually work

[lethal apex]

First half of this sentence is correct, second half I’m not so sure about. Wdym by copy when you reassign?

[low vault]

LOL I was too bricked to do real modding

Plugins were easier to setup

[low vault]

Like

If you have a list L, in order to copy it you have to do like M = copy.deepcopy(L) rather than M = L

By reassign I meant assigning it to a different variable name

[quasi orbit]

i thought this was like

how ur supposed to do it

lol

its only cpp thats weird

[lethal apex]

Yes that’s correct, idk where you got reassign from lmao

[stone vector]

what quant are still hiring

[low vault]

Bad English ig

Oopsie

[low vault]

Idk I find the other lands weird

*langs

I guess it’s cos I view assignment as more of an explicit copy operation vs aliasing

[lethal apex]

^ agreed, I want full control over whether I pass by value or ref, and by default I want b = a to create a copy of a. Python does a ton of these choices under the hood and without my permission which is why I would never use it for anything serious.

[stone vector]

i use it for ML

optimized library

[lethal apex]

Yes, although the trivia specifically was the fact that python caches integers from -5 to 256

[low vault]

:p

tbh I disagree with how C++ handles it either -- I find the default behaviour more intuitive, but C++'s behaviour is way more inconsistent because the cases of when the different operators are chosen is super unclear

disallowing copying of objects on assignment is the easiest and most consistent way to avoid this issue IMO

I also agree that enterprise Python is a crapshoot, but rather because of the lack of type checking (type hints are not strict, and many old codebases don't have them)

I think the way it defines = is fine

[lethal apex]

do you have an example for this

[tranquil wagon]

I guess for people who are unfamiliar with the language

MyClass obj = otherObj;

Seems like it might call copy assignment operator but it calls copy constructor.

Also initializer lists were such a bad thing to introduce to the language, I truly despise them

[tranquil wagon]

I agree that there’s a bit more to learn up front to write good c++ but “inconsistent” definitely isn’t the right word, the choice of which assignment/copy function is called is extremely well defined and once you know the rules you can tell immediately

[zenith panther]

one off knowledge aint useful

it shouldnt be how is x implemneted

it should be

how would you implement x

[zenith panther]

why???

its faster

[lethal apex]

Disagree as long as you mention there’ll be trivia during the interview

[zenith panther]

nah

[lethal apex]

I’m not being totally serious but you could also weigh the trivia segment close to 0 but use it to see how candidates react when they don’t know something.

[zenith panther]

i strongly disagree

thats just stupid

[lethal apex]

Tc?

[tranquil wagon]

It’s not faster if you know how to write a constructor properly

[zenith panther]

im a university student

[zenith panther]

really?

how?

u do it in-place right

interesting ill take a look at this

is the problem with initializer lists fhat u dont like the syntax or u fundamentally disagree with the idea

[tranquil wagon]

Just don’t let the members get default initialized

[tranquil wagon]

The syntax

[zenith panther]

so ur against RAII?

[tranquil wagon]

What? No

[zenith panther]

ih

so u just want it to be garbage

until initialized

i c

[zenith panther]

well then how r u gonna destruct garbage

[tranquil wagon]

No just write a constructor that takes all the members as arguments and use the member initialization syntax instead of letting them get default initialized and copy assigning them

[zenith panther]

eh idk

[tranquil wagon]

Then there’s no performance difference from an init list

[zenith panther]

i think init list is good

u have to add addtl logic

[tranquil wagon]

The syntax for initializer lists causes such unreadable code

Exceptionally error prone code

[zenith panther]

i mean maybe? cpp isnt an elegant language

the alternative u propose would probably break raii

[tranquil wagon]

It actually is a lot of the time

[tranquil wagon]

No it doesn’t

[zenith panther]

or introduce more footguns related to raii

[tranquil wagon]

I don’t think you know what you’re talking about

[zenith panther]

i think i do

[tranquil wagon]

I’m not talking about adding a language feature

It’s just if you write good constructors then initialzier lists aren’t more performant

[tranquil wagon]

I think most of the time it is

[zenith panther]

idgi, dont u have to do an additional copy if u assign unless u have it be in the like

default value

like youll make a stack version and then copy into the member

[zenith panther]

ehh

[zenith panther]

could h provide an example i rly dont get what ur saying

[tranquil wagon]

Yes one sec

[zenith panther]

maybe i am misunderstanding heavily

are you saying we should have a compiler error for when we dont assign a value on ctor

[tranquil wagon]

struct MyStruct
{
  MemType m1, m2, m3;

  // bad, init list will be better:
  MyStruct(const MemType& arg1, const MemType& arg2, const MemType& arg3)
  {
    m1 = arg1; m2 = arg2; m3 = arg3;
  }

  // good, init list will be same
  MyStruct(const MemType& arg1, const MemType& arg2, const MemType& arg3)
    : m1(arg1), m2(arg2), m3(arg3) {}
  MyStruct(MemType&& arg1, MemType&& arg2, MemType&& arg3)
    : m1(std::move(arg1)), m2(std::move(arg2)), m3(std::move(arg3)) {}
}

[zenith panther]

arent these all init lists

like the two under good

[tranquil wagon]

do you see std::initializer_list<T> anywhere

[zenith panther]

() : m1() is also called initializer list as well

[tranquil wagon]

no its not

[zenith panther]

https://en.cppreference.com/w/cpp/language/constructor

member initializer list

[tranquil wagon]

oh wait weird

thats not what most people are referencing when they say initializer lists

[zenith panther]

two diff things

but yeah

this is why i was confused

[tranquil wagon]

https://en.cppreference.com/w/cpp/utility/initializer_list

this is what I'm talking about

[zenith panther]

yeah

this is why i say cpp is not elegant

bcz of BS like this

also coroutines…

[tranquil wagon]

these are the worst

i hate these so much

so so so os os o cnuh

much

[zenith panther]

variadic function list is even stupider

its so fucking ugly

just parameter pack 😡

ok not actually

but its still ugly

[tranquil wagon]

these are fine

[zenith panther]

no it is ugly

[tranquil wagon]

my issue with initializer lists is that any time braced initialization is used, type deduction says its an initializer list

[zenith panther]

and i like std::initializer_list for creating a vector sometimes

[zenith panther]

oh waz

it is usually explicit ctor or smth

weird

[tranquil wagon]

• with auto type deduction

not with template type deduction

[tranquil wagon]

what do oyu mean

[stark juniper]

What's a monad

[tranquil wagon]

To elaborate on why the type deduction with braced initialization is annoying to me, look at this example:

std::vector vec{other_vec.begin(), other_vec.end()}

[stark juniper]

What's a mono

[tranquil wagon]

Common sense said this copies the other vector, which it would if you used parentheses instead of braces

[fresh swallow]

Lol I was reading this conversation mad confused

[tranquil wagon]

But instead it calls the initializer list constructor, and makes a vector of std::vector<std::vector<T>::initializer> objects

[fresh swallow]

Cause that's what I think of when someone says init list

[zenith panther]

oh 💀💀💀

what the fuck

[tranquil wagon]

Yeah initializer lists are the worst

[zenith panther]

u think of the ctor too

yeah

ig u only rly use {} as explicit

bery odd tho

[fresh swallow]

This is nonsense code though

[tranquil wagon]

what do you mean by this

[fresh swallow]

Who writes vectors without the type

[zenith panther]

only use it wuen its only kne parameter

[tranquil wagon]

no its not, it could easily be written by someone who doesnt know the language well

[zenith panther]

and u mark ctor as explicit so jt wont get casted automatically

[fresh swallow]

Why would someone who doesn't know the language well be writing it

[tranquil wagon]

thats a language feature so it is intended to be used

[tranquil wagon]

why would you ever learn a new language

[zenith panther]

well cpp is one of those languages not meant for these ppl unfortunately

[fresh swallow]

I wouldn't

I know c++

[zenith panther]

cpp only real use case rhese days is performance engineering

[tranquil wagon]

everyone was one of these people at some point

and it makes the language harder to understand

[tranquil wagon]

until today you didn't know this would call the initializer list constructor

[fresh swallow]

What edition of c++ introduced the ability to do this?

[fresh swallow]

I did know that 🤔

[zenith panther]

it is quire odd

[tranquil wagon]

17 iirc

[zenith panther]

but if u reas the standard it sint too confusing

[tranquil wagon]

not you, him

[zenith panther]

actually

[fresh swallow]

Most people learn c++11 or something

First

In school

[zenith panther]

or refernexe ig

[tranquil wagon]

seems like a weird decision

[fresh swallow]

Curriculums change slowly

[tranquil wagon]

Ok, in 10 years maybe they'll learn c++ 17 and by confused by this then

[fresh swallow]

Maybe 🙂

[tranquil wagon]

I don't see your point

Just because your school is teaching you old C++ doesn't mean everyone learns old c++

[zenith panther]

the real villain was actually not specifying the template for vector

[fresh swallow]

My point is that it's dumb to say "This feature is bad and shouldn't have been added" and your reason is a piece of code no one would ever write

[zenith panther]

if im gonna be honest

[tranquil wagon]

You might have thought the type deduction was clear in that case

[zenith panther]

• first compile would fix this issue

[tranquil wagon]

it wouldnt be a compile error

[zenith panther]

it will when u do antthing with the vector

[tranquil wagon]

Are you trying to argue that this was a good design decision?

And it shouldnt call the vector iterator constructor?

or are you just being contrarian

[zenith panther]

im saying that no one would ever run into this issue

and not have it fixed in like 5 minutes

[fresh swallow]

I additionally think that there are scenarios in which initializer lists are valuable

[tranquil wagon]

Ok, can you explain to me the differences between:

using IntMap = std::unordered_map<int, int>;
IntMap mymap({{{0, 0}}});
IntMap mymap{{{0, 0}}};
IntMap mymap({{0, 0}});
IntMap mymap{{0, 0}};
IntMap mymap({0, 0});

[fresh swallow]

Like if you're working with generic code where you can have some unknown number of types

[tranquil wagon]

You even have the full type here so the map template arguments arent even being deduced!

should be easy right

This is what braced initialization leads to

[fresh swallow]

Trivia questions?

[tranquil wagon]

If you saw all of these next to each other in a codebase are you telling me you would immediately know what the difference between them is?

without braced initialization it would be easy

with braced initialization its hellish

That's why I dont like it

[lucid cloak]

I think bro might be good at cpp trivia

[tranquil wagon]

this isnt trivia, this is a question i got from someone who was confused while writign a project

[tranquil wagon]

If you think the language would be better without each of these meaning something different (and this is without the type arguments even being deduced) then you agree with me that braced initialization was a bad design choice

[zenith panther]

i would just compile them and see what it does

waz

[tranquil wagon]

ok bro

[zenith panther]

and also theres just good and bad code

like if u use x feature in a bad way

obv it is bad

[tranquil wagon]

obviously

[zenith panther]

just bcz u can do smth bad doesnt mean u should

[fresh swallow]

They were confused because they wrote these and then forgot why ?

[tranquil wagon]

you are just saying truisms like that suppoorts your argument

[tranquil wagon]

They didn't know which they should use

[zenith panther]

what is my argument

[fresh swallow]

Why not

What did they want to do?

[tranquil wagon]

I don't know if you have one, thats why i asked earlier if you're just being contrarian and disagreeing with everything i say lol

[fresh swallow]

It is difficult to look at 5 different things where I don't know why anyone would write them and say exactly what they're doing yes

It's not that hard to use them properly when you are writing real code

[tranquil wagon]

would you be able to explain what each one does

without referencing the standard

[zenith panther]

exactly

in my experience ive never run into this

and also irl u have the standard and reference for a reason

[tranquil wagon]

Yes but the best way to use it properly is to NOT use it imo, and it is even easier to use it badly than it is to use it well

[zenith panther]

this is exactly why i hate trivia questions

theyre stupid

[tranquil wagon]

Go write c++ professionally for a bit then come back and tell me that braced initialization should stay

[fresh swallow]

I have been writing c++ professionally for 3 years

[tranquil wagon]

You think braced initialization is on balance good?

[fresh swallow]

not once have I run into an issue where I didn't know what an initialization of something did

[zenith panther]

i did

i wrote it at citadel

[tranquil wagon]

And do you think braced initialization is good?

[fresh swallow]

I don't think it's bad

[zenith panther]

i like using it for vectors and stuff

[fresh swallow]

And I don't think it would have been added if it didn't have valuable use cases

[zenith panther]

idt its that bad haha

[fresh swallow]

Plus yes, it makes creating stuff like vectors and arrays in one line possible

[tranquil wagon]

That answer is literally applicable to everything in the standard

Do you have any opinions of your own

[fresh swallow]

Not really

The language is a tool which I use to get the job done

I can always manage to get it to do what I want because of the number of options available

I am against the idea that there should be less options because some people somewhere might not use them properly

Given that my coworkers are in my experience competent and do not write trivia question esque code

[zenith panther]

that cpp coroutines were horribly designed

[zenith panther]

mhm

trivia is stupid

[tranquil wagon]

Anyone who is competent in any language can write good code in that language, there's more to consider when talking about language features than "if someone gets as good as they can get at this will they be able to write good code in it"

Some language features have such limited upside with such easy to run into downside that I don't think they should exist

[fresh swallow]

"Limited upside"

[tranquil wagon]

Yes

[fresh swallow]

How should we create a vector with existing members without initializer lists?

[tranquil wagon]

Everything is a tradeoff

[tranquil wagon]

reserve and push back

[fresh swallow]

jesus

[tranquil wagon]

or variadic templates

everything is a tradeoff

[fresh swallow]

You think we should have to do that just so that people can't write stupid initializations?

[zenith panther]

no…

no…..

[tranquil wagon]

Ok variadic templates then

[fresh swallow]

I know you're a removing monkey bars from the playground type guy

[tranquil wagon]

Or we can just fix the type deduction rules around braced initialization without removing initializer lists entirely

[zenith panther]

variadic templates… binary bloat to infinity…

please no

[tranquil wagon]

how about this then

All I'm saying is the way they work right now is not desirable to anyone

[zenith panther]

maybe

i only use braced init for single parameter ctor

so i have no idea

[tranquil wagon]

That use can be trivially replaced

with a regular constructor

[zenith panther]

it looks cooler

🤣

but yeah maybe

no idea tbh

i havent seen the upsides

ive inly heard the negatvie part from u (which mostly sounds like a skill issue)

[tranquil wagon]

You say you have worked with c++ professionally, if there were valuable upsides you'd think you'd have come across them yourself

[zenith panther]

no it just means i never used it

not everyone uses cpp coroutines

not everyone uses tmp

[tranquil wagon]

How have you never used it, they are integral to the language

[zenith panther]

not everyone uses a bunch of stuff

[zenith panther]

idk, ig it wasnt integral enough for me

😭😭😭

[fresh swallow]

Just cause a feature is used when writing compilers and standard library things doesn't mean your average dev needs to use it

[zenith panther]

yurp

im library writing rn

holy moly

library writers gotta do so much

[fresh swallow]

And the average dev not needing to use it doesn't make it not important for building low level components

[zenith panther]

mhm

[tranquil wagon]

I build a lot of low level components.

[zenith panther]

mr big chest understands it

ok

[dire finch]

this seems unnecessary

[tranquil wagon]

It wasn’t the best suggestion

I’m fine with keeping initializer lists honestly

Just the type deduction rules suck as they are rn

[floral salmon]

I build a lot of low level components.

[dire finch]

I segfault a lot of low level components.

[low charm]

anyone else get 2 sigma freshman swe OA?

[thin quest]

Orz…

[rapid phoenix]

Any idea, what is the future like for a risk developer at hedge fund in these AI times?

[neon mirage]

How do I avoid checking every process to efficiently find the ones that terminated abnormally? In C. Ik it may be dumb question

[minor wadi]

Yea I got it a few hrs ago

How ru gonna prepare

[ruby ravine]

Median SWE is probably gone some time in 2026, finance / trading in particular is probably longer because most of those codebases are not public so models will probably be worse than humans for longer. But honestly as a human we’re all in the same boat, and the rising tide comes for us all

[unkempt oxide]

how did your interview with them go? i got reached out by them

[pulsar scarab]

Rejected

[unkempt oxide]

was it for SDE or quant trader?

[pulsar scarab]

I don't even remember lmao

[unkempt oxide]

😭

can i dm u

[pulsar scarab]

Swe?

I think

Cuz we did coding

[unkempt oxide]

i wanna see if job description is same

[stone vector]

if median swe is gone and quant swe won't be because codebase isn't public

then quant firm can just train the thing on their codebase

[daring basalt]

bro AI is so bad at programming

If your job as a programmer is to do really basic functions that an AI can do yes you're fucked otherwise it's nowhere near close enough to worry

[dire arrow]

cyap

keep coping gang

[slate crown]

qswe will be over too lmao, it’s not like AI isn’t trained for low level stuff + internal code at the firms

[lethal apex]

“Low level stuff” - ask any senior low latency engineer why they make decisions. Or worse, a hardware engineer. Half of the time they’ll give a non-answer like “intuition.”

[low vault]

[lethal apex]

Kind of reminds me of AlphaGo where the top go players can’t even explain their logic, so I’m not entirely skeptical.

[low vault]

most thinking is unconscious anyway so unsurprising tbh

*subconscious

[lethal apex]

Unconscious lol

[low vault]

many codebases are pretty old and don't have those features / integrated

[lethal apex]

Walking into your firm to find the whole floor asleep

[low vault]

honestly ty for the example this is pretty much exactly what I was trying to think of

[tranquil wagon]

Always happy to give random examples of c++ lol

[tranquil wagon]

Also if u want a fun exercise try looking at the standard for unordered map constructors and matching these to which constructor it will call

(Even more fun, one of them won’t compile at all)

[lethal apex]

on serious note the main point I’m making that either one of two things needs to happen for some of the high-skill engineering disciplines to be automated:

1. Thousands of hours of interviews from senior engineers as well as them vocalizing their decisions as they work.
2. Agents need to be able to simulate different hardware/cpu states and interact with completely realistic computer environment. (Inspired by alphago simulating Go boards except instead of 361 squares you have a thousand.)

[low vault]

fair enough that it's not actually inconsistent. But I still think that the way the code is layed out is hard to read and error prone -- going back to your example, while it's unambiguous what the line of code means, and ideally the line of code should be well-defined regardless of whether the copy ctor or copy assignment are called, I think mistakes are still possible (and would be a massive headache to debug). I think if I'm working on a code base that is shared across many people, I'd like the language itself to have minimal footguns so I don't have to go around plugging peoples silly mistakes (and vice versa for my mistakes). While a more restrictive language makes code harder to write in the short term, I think it's avoiding pain in the long term.

I think C++ doesn't make it clearly visible which choice is made (even though it is consistent)

[low vault]

😭

thank god for library writers that will define all of the ctor variants so I don't have to lololol

[thin quest]

bruh when will drw reachout

[tranquil wagon]

Tbf most of the time your copy constructor and copy assignment operator are functionally equivalent, you don’t really need to think about which is being called in that example (and you can just be happy the compiler knows to link the constructor so you can avoid an unnecessary copy), but yes I agree that that there are some footguns in c++ like braced initialization stuff that the language would be better without

[tranquil wagon]

They already finalized intern class for this summer

[thin quest]

discover drw

[tranquil wagon]

Also if you’re just using the language casually and your classes just have members that are built in types/types made with raii principles by someone else then you can just trust that the compiler generated ctors/assingment operators will work without even having to define them yourself

[low vault]

yeah casually I don't think there's any problem with C++

I only really used it for CP before internships lol

though arguably almost every language is fine to use casually

[low vault]

fair enough. I wonder if there's a case where someone made a bug in the copy ctor but not the copy assignment, or something like that

I feel like I can definitely forsee people missing one case of the ctor and forgetting about it until later on

like sure it's skill issue but I've seen people make silly mistakes even at big tech firms like nvidia

and also really jank codebases at shaw (though not in C++)

[tranquil wagon]

There’s this whole thing called the rule of 3 and rule of 5 that says which functions the compiler will generate based on which ones you manually defined (not sure if that’s what you’re talking about) https://en.cppreference.com/w/cpp/language/rule_of_three

It’s essentially this

[low vault]

ye

[tranquil wagon]

Yeah honestly one of the hardest things to do when I’m making a large project is make sure all my ctors and assignment operators are done correctly lol

[low vault]

I still find it unnecessarily janky though, since it's something that's very C++ specific (and so you really need to be locked into the system to know it), and idk if I trust people knowing it off the top of their head when it's necessary

[tranquil wagon]

Especially when classes manage resources that are stored on the gpu

[low vault]

I don't think you end up declaring new classes very often either

i rmb learning rule of 5 in OOP class and then promptly forgetting it like a year later lol

I remember most of it but I think you really need ti locked in to be effective

[tranquil wagon]

I think u go through phases of declaring a ton of classroom then not many

Then a ton more again

[low vault]

yeah

this is why we should all migrate to zig clearly

[tranquil wagon]

Lmao

[low vault]

or D or something like that

AFAIK rust is too rigid to work with often

[tranquil wagon]

But yeah for things where you have a class managing some buffer on the gpu you really want the ability to manually define move/copy ctors and have the behavior be different

[tranquil wagon]

I can confirm based on the last week of work I’ve been doing lmao

I hate that language so much

[low vault]

blobspin

i tried doing some programming problems in rust for fun and it got too frustrating at points tbh

[tranquil wagon]

It’s just too idealistic and you end up needing to do everything in unsafe blocks, especially when you need interop with c (which you almost always do, because all mecessary third party libraries have c apis)

[low vault]

mmm right

tbh mostly I think I just wish C++ had a syntax change to be more clear on when something is a move/copy/etc.

but honestly outside of CP the only experience w c++ I have is from internships so I don't have much to say off the top of my head

[tranquil wagon]

I wish it was less clear (crying in std::move rn)

[tranquil wagon]

Internships/jobs is probably the best c++ experience you can get tho because u see what’s useful in real life

[low vault]

🥺

most of the C++ I write just feels like glorified OOP code with extra steps tho LMAO

and every once in a while i get slapped by some os-specific detail and shit myself

[low vault]

yeah std::move is jank as hell syntax-wise

needa sleep so will log off for now but ill reply to stuff in the morning :p

[tranquil wagon]

C++ is used for OOP most of the time tbf, it has a lot of nice features in its classes

[tranquil wagon]

Gn!

[thin quest]

https://qr.link/3tZ6NC

doesn't work

[ocean gazelle]

deadline was 1/22

[unkempt oxide]

it also just has more modern features like smart pointers etc

[zenith panther]

yeah, rust is not the best for performance bobbing imo

[zenith panther]

the guy i work on rs with thinks that D is goated

he did contribute to the project so ig thats part of why

[ruby ravine]

idt one codebase is enough here, it’s not just “see one codebase, know everything about it,” you need a lot of high quality examples

[ruby ravine]

I dunno, perhaps you are a much stronger programmer than me but AI can do ~80% of my job already and this number is increasing

Feels more likely to me that you’re just not using the tools that are out there / using them properly

[ruby ravine]

1. OpenAI is literally paying PhD students to do nothing except label training data 2) OpenAI operator is pretty clearly trained using reinforcement learning on a realistic computer environment

[stone vector]

idk personal opinion but if an AI become good enough to replace regular swe it'll prob get good enough quickly enough to replace quant swe

high quality examples aren't enough to replace regular swes also is why I said that. once an AI become good enough to learn on its own through trial and error with reinforcement learning rather than relying on high quality examples that's when it will be good enough to replace regular swe

and once it becomes good enough to learn on its own through reinforcement learning then it can also replace quant swe

but who knows

[lethal apex]

Neither of those apply to what I was talking about though. Top industry professionals gained their expertise through hours of experimentation and watching entire product cycles. Operator is not simulating cpu internals it’s interacting with a browser.

The problem with low latency is that it has low training data and at the same time requires more reasoning.

[ruby ravine]

you don’t simulate CPU internals either, but you interact with an interface that lets you interact with them under the hood. browser / no browser doesn’t matter, you’re letting an LLM use a computer

phd students / top industry professional is all the same stuff, the point is you’re paying smart humans for reasoning traces

[ruby ravine]

why do you think high quality examples are not enough?

[lethal apex]

“all the same stuff” no and I’m almost convinced you are baiting. The gap between a grad student and an industry veteran is enormous.

[ruby ravine]

RL is pretty dogshit, like you end up wasting a ton of compute on exploration, you want to do SL if you can

[ruby ravine]

not really?

and like even if we disagree on this, scaling labs are still paying money for high quality reasoning traces from specialized people

who the scaling labs pay money isn’t really that relevant?

like if they think they get more bang from their buck from people in industry I’m sure they would

[lethal apex]

I was offering a solution as to how OpenAI could train a model to perform low latency/hardware engineering. The options are you either collect enough training data, even the intuitive parts that veterans don’t share, or you try to discover these on your own. That requires simulating internal hardware. Maybe an argument could be made that all you need are profiling tools, in which case I’d agree for some low latency work.

[ruby ravine]

I think all you need is something like operator just with a full OS?

It depends on how much you believe in RL

[lethal apex]

Not really a good assumption, the number of grad student compared to industry veterans is incredible. The reality is that PhDs will certainly help in some areas but I doubt it will be enough.

[ruby ravine]

What is an industry veteran?

Like would you consider an L5 at Google an industry veteran

[lethal apex]

Someone who has consistent and lengthy experience in the field of low latency/chip design.

I’m specifically talking about this field and not all software.

[ruby ravine]

Sure I mean like it’s a subset of software that happens to have less training data

[lethal apex]

compared to a browser this requires exponentially more compute to run through simulations. When you factor in non determinism and a constantly changing landscape, it’s really a different problem.

[ruby ravine]

I don’t think low latency programming is particularly hard compared to regular programming

[stone vector]

i just don't think it is

[lethal apex]

Yea I disagree, it’s one of those fields where the “training data” for humans is largely experimental.

profiling non deterministic data sucks

but yea I would say getting efficient os simulations should be a #1 priority to achieve a low latency agent

[lethal apex]

you funny

do me a favor and connect w an intel manager on linkedin, just repeat what you said to them

they've needed a good laugh recently

[zenith panther]

programming is only as hard as you make it

if you push yourself eventually you will reach tue same difficulty regardless of what you put your pts into

[ruby ravine]

eh i dont really see programming as some skill that you have to hone to the limit, there's diminishing returns to everything so you just hone it to the extent that its a good decision economically

i think the point where it stops being a good decision is before you start doing the actually insane / difficult shit

[zenith panther]

maybe, but eventually youll find other routes to increase benefit

[ruby ravine]

ah are you referring to me or someone else above?

[zenith panther]

why cant the agent just run their code and profile it like ppl do rn?

or do u mean like at the hardware level

[slate crown]

yea i agree w u, i didn’t mean rn rn but def in 5-6 years

[lethal apex]

No I meant rift

[lethal apex]

That is also a possible solution, yes. But the key here is that we shifted from “this ai can be trained to be a good engineer” to “this agent needs to explore its domain by profiling.”

[zenith panther]

but its an agent right

[lethal apex]

The reason I suggested simulating the OS/cpu is because understanding the hardware/kernel level is a far more difficult thing than C++ code generation

[zenith panther]

the pt if y give it the ability yto interact with stuff

[lethal apex]

The reason we’ve been able to scale LLMs so well is because we aren’t simulating dialogue, we converted it to a simpler task: token completion

[zenith panther]

hm. ig maybe if u have custom fpga setup itll be harder.

[lethal apex]

The number one challenge facing ML today is making efficient, parallelizable, and realistic simulations. Do that and everything I said is wrong.

(And we usher in a new era of robotics, chip design, physics research, etc.)

[zenith panther]

jm

i have no opinions

¯_(ツ)_/¯

[lethal apex]

Wdym

[zenith panther]

not my domain

[zenith panther]

i imagine perf engineering would be p easy

but if u have custom hardware the llm wont be able to work w/ it as easy

since it doesnt understand the hw

i imagine ppl will figure out how tk make it workt jo

or maybe AI will

🤣

[lethal apex]

Well the point I’m trying to make there is that humans also have to do their own simulations, and I’m not convinced an agent can simulate low level better than humans yet.

[zenith panther]

oh yeah

well it cant

bcz otherwise we wouldnt have low level engineers

its more abt is it realistic given what weve seen aand future progress imo

[lethal apex]

Funny thought is that an Ai engineer would profile at about the same speed as a human

because if you could simulate the entire OS faster we’d just use that OS 😂

[zenith panther]

it could probably generate more prototypes than a human could tho

so itd be able to profile in parallel

wait actually holy joever

maybe AI will be better at this stuff than frotnend dev

holy moly

guys its over

[dire arrow]

u do realize the answer to that is highly dependent on ur interests and goals right

[lethal apex]

max dama's blog is infamous now

[near nymph]

like making urself the chad wojak

[ruby fog]

yo

[last rune]

noobie here but can someone without stat/math concentration get into quant within just one year by just spamming codeforces

or is there more to that

[neon mirage]

For SWE? not probable unless you have strong prev exp and assuming you are a CS major

[last rune]

ye assuming cs lol

like I dont have stat or math minor at all just pure cs

how long is this book

[neon mirage]

are you at a good school CS wise? Do you have prev exp? Are you starting from 0 codeforces/leetcode exp?

[last rune]

t20

junior but doing masters

[neon mirage]

oh then its possible

[last rune]

hmm ok

[thin quest]

what's roadmap

[neon mirage]

do you have any prev intern exp?

[last rune]

yea

[neon mirage]

its definetly possble