my work ethic 5xd when I stopped telling ppl i would do something til after i did it
yapping abt future actions kills motivation to execute
if u say "im gonna build xyz!! hey boss im gonna do abc this week!!" you will have received your reward, wrecking the incentive to work
pro tip for learning in new fields
every word you hear, and don't know, LOOK IT UP
EVERY WORD. takes time but eventually you know all the words
if there's too many new words, settle for looking up every word you hear more than once
pro strat: get tons and tons of skills by building stuff
never made a site w stripe payment? do it to learn it
never made an image classifier model? DO IIT
skills are super useful, arent taxed, cant be stolen/lost, AND THEY COMPOUND bc they synergize
This is rly mostly bullshit
Many friends hired who have no advanced degrees but they can code in Python and know how to work with AI models
Don't believe in credentialism, believe in skills!
for your day to feel super productive, do tasks in the following order
1) time sensitive
2) tasks you want to do the least
3) tasks you like to do
if u do work you like and avoid work you dont, its just procrastination. feels bad
to feel amazing, do what you like least, first
I HIGHLY HIGHLY HIGHLY recommend trying this. after testing it for 2mo, it consistently makes my day WAY more productive/rewarding
definitely will be a lifelong habit for me, along with
- start things immediately + spread the work out
- get good sleep
for your day to feel super productive, do tasks in the following order
1) time sensitive
2) tasks you want to do the least
3) tasks you like to do
if u do work you like and avoid work you dont, its just procrastination. feels bad
to feel amazing, do what you like least, first
@OrphicCapital
When i was a little kid it was raining and worms were everywhere, vibin
Ppl were stepping on them and i didnt like that so i told everyone it was bad luck to step on worms and they stopped
article on how to get cracked at building things quickly
funny experiment: DONT follow or RT UNLESS you try this and it works for you. but if it does, u have to come back and do that
btw i only write abt stuff that has worked for me for months or yrs
how to learn by building, from 0 to cracked:
1 learn basics w flashcards/online tests/tutorials
2 find an idea for a project slightly outside your skillset
3 build it, forcing you to increase your skillset in order to finish
4 repeat 2 and 3 forever
recap of my articles about useful stuff:
how i cured my procrastination
how to build anything extremely quickly
how to max out your knowledge derivative
cracked LLM techniques
To succeed in most things, focus on the fundamentals (the real). I.e. get CRACKED at the base concepts that ripple out to everything else. 80 20 kinda
The middlewit "its about tactics/tricks/hacks" view will hold you back. I had it for yrs its TRASH
People so steeped in fake shit they’ve fully lost sight of the real. Think everyone is faking it, real is a children’s-tale or myth, etc. Shock and consternation at witnessing real, like meeting an alien - worldview blown
highly hiiiiiighly recommend doubling down on making your own tools
made one today that trims a 15s process (300x a day) into 1s
and another that cut a 3min process (1-2x/day) into 1s
feels soooo soo SO good man. like building an army of tiny little robots
massive list of all the cracked LLM techniques/tools I personally have found to be super useful
includes examples, images, "copy prompt" buttons so you can easily try em
16hr coding session
#11
progress
we do 16hrs of work every mon and thurs. join!
ilya30u30 papers (supposedly what ilya gave carmack?)
hiiighly recommend making your own tools like DANG dude they speed things UP
I just want to find 100x or 1000x engineers
surround myself with them, and become one
you become the average of the people you spend the most time with
elon is a 40,000x engineer, im sure there are lots of 100xers out there
the reason I keep saying "you need to do more volume" is because intelligent people win at literally anything once they recognize the patterns
I have never once failed at anything I put my mind to, because I eventually figure out every system and its loopholes
my method looks
@allgarbled
Doesnt this just make performance feedback a weaker signal
Is the meta then to be much more cautious abt performance feedback you hear abt friendly ppl
16hr session done. feelsgooooooood
we do these every mon and thurs. key is consistency. work your way up to 16. you want to do this long term and enjoy the progress
compression book:
ilya30u30:
despite the clickbaity title, this video has really good advice (imo) on how to get better at programming
basically: understand your tools and libraries, rebuild stuff from scratch in c to learn, and make sure you have fun
if you wanna get cracked build stuff 15% out of your skillset, repeatedly. its fun too
learn by doing is super super powerful. comfort zone is the enemy
each proj grows skills by ~15%, doubles your skillset every 5 projects. speedrun this
been doing this over half my life now
AMAZING 16hr session, got a ton done
data compression explained:
ilya30u30 papers:
my brief ilya30u30 notes:
the discord is getting more cracked by the day I swear. so so cool
if you teach people to be cracked + 'pay it forward', would that spread like a virus?
let x = teach ppl to 1) be cracked and 2) x
like could we engineer a good mind virus that makes ppl cracked at scale?
Angkul is a great lowbie to invest in early
Hes building his own web browser from scratch for fun, and constantly grinds learning ML and programming. Its been fun seeing this dude get more and more skilled by the day
Code report: Day 48
16 hr code session done. work around 11 hrs.
score: 7/10
> work on implementing the js in web browser(chapter 9)
> build two linear regression model from scratch
> lecture 3 of cs231n. karpathy sensei's slides are best.
cya ;)
some signals are noise
to be cracked at anything, find and get cracked at the fundamentals (this is recursive)
havent u realized the whole reason they're taught first is bc theyre the most useful?
my life got way more interesting when I realized that going above and beyond (in the right circumstances) can bring you incredible rewards and amazing, meaningful adventures
How to overcome plateaus. goes over one of the strongest learning techniques that has helped me.
(it also has chess in it bc I love chess)
tldr: overcome plateaus by improving via finding (and then training) core concepts you’re currently overlooking
Sleep has such a profound difference on your ability to write and think
The entire past month ive had a dropoff in ability to write articles, code, post bc caffeine was masking my sleep debt
Took a break, slept well for a few nights, BAM much more energy and thoughts flowing
16hr session complete
i am so tired but it feels so good
had to go an extra hr cause i got distracted by chess
(i am CONVINCED visualization practice is how you get super strong)
btw we do these sessions every mon and thurs, heres the discord
if you put energy into what you do, and try to never lie (its hard), you set yourself up for an adventure
maybe thats why ppl are so nihilistic, we've all been getting sniped by some dumb middlewit grifty tactics mindset that kills the call to adventure
16 HR WORK/CODE/STUDY SESSION
#17
(+/- 1)
COMPLETE
felt super productive this time, minimized distractions for the most part
join the discord we do a session every mon and thurs!
metacognition will make you cracked at solving your own problems
you get skilled at understanding your own thoughts by practicing noticing them for yrs
i talk about this a bit here
Im so so hyped for ding season 2
My favorite posts are seeing people build cool stuff
Pro tip, like+bookmark every cool post like this for x to fill your feed with them. I do this and my FYP is filled with the coolest shit
fyi if you drink lots of caffeine it can mask how sleep deprived you are. youll feel well slept but your brain functions worse
it was happening to me over the past month, i couldnt figure out why I couldn't think as clearly
taking a day off caffeine helped fix this
if youre stressed bc you work/learn slow, assume this razor:
you arent dumb. you use bad techniques/algorithms for learning, working, etc. fix them to be smarter/more efficient
this is VERY VERY VERY common. ive seen ppl 1.3x, 2x, 5x from fixing algos
Having a strong why is SUUPER useful, and the largest disadvantage I got from lack of sleep last month, tbh, was forgetting the 'why' behind my goals
Why do i post? Why study programming? Why write articles? Whats the point?
Slept. now i remember :)
Sleep has such a profound difference on your ability to write and think
The entire past month ive had a dropoff in ability to write articles, code, post bc caffeine was masking my sleep debt
Took a break, slept well for a few nights, BAM much more energy and thoughts flowing
To speedrun learning a complex concept, find the most commonly talked about subconcepts. Then learn those. do this recursively if needed
practice/get good at this
for transformers, some fundamentals are: self-attention, residual nets, encoder-decoders
@dnbt777
what's your approach for adapting your speedrunning outline technique to concepts that you have little to no knowledge of? i.e. in studying ML, if I want to implement a simple transformer, but don't even know what I don't know, it's challenging to create even a rough
ok, as much as I hate this, no more new projects until I finish more old ones
shiny object syndrome is real man
way better to have 4 complete projects than 20 half done ones
great account to follow for ML/reinforcement learning
@jsuarez5341
super skilled MIT Phd
great stream so far, ive learned a lot, he took snake from 200steps/sec to like 500k
here are ludwigs ML/CS resources as a RAG u can chat with:
includes 201 YT vid transcripts, 88 bookmarks, 1 book
setup:
1 download/unzip/open the obsidian vault
2 install smart connections obsidian plugin
3 add your gpt api key
implemented lz78 compression algo. quick fun exercise, i recommend it
got the idea from
@pixqc
(great ML follow imo)
resources:
clear, concise notes on lz78 by Peter Shor (of Shor's algo):
random cracked russian guy on YT:
@yacineMTB
x is modern athens
you have the ppl that make you think. you have diogenes. you have it all
used to be universities but they fumbled it. flipped it and tumbled it
@nearcyan
fear of regret of not achieving my life goals made me put tons of willpower into turning my entire life around. it worked
if you grew up thinking youll be successful, it makes it 100x more painful to give up
not saying the trait guarantees success, but there is a causal factor
16HR SESSION
#13
, COMPLETE
links below to discord, ilya papers, compression book, my ilya paper notes
JOIN its a VIIBE + we get stuff DONE
we also share our learning resources, and our tips on what helps/doesnt for working long hrs
i used to play vidya for like 14hrs a day on my off days
now i just grind work or side studying/projects
its the same thing to me except this feels 100x better fug
16hr session COMPLEEEEETE
join the discord we typically do one every mon-thurs
great opportunity to get stuff done and improve your work ethic with others in CS/SWE/ML who are serious about it
Had a GREAT 16hr session today. First one after ~2 weeks or so (i was away and unable to do them)
From now on I'm only posting my progress in the discord bc I want to experiment with making my x a place for posting 1) useful information, and 2) completed results, and thats IT. I
@rtx140201
practice the habit of starting everything immediately. Then spread the work out over a long period of time, preferably finishing early. Finish each work session hungry for more by stopping slightly before youre satisfied
If you do one thing, do this
I went from C/Fs to As w it
Amdahl's law
it's why getting cracked at the fundamentals is the route to mastery in anything, since they're used the most
"Black belt is the belt of having a firm, solid grip on your basics" - Roy Harris, 7th degree bjj black belt, badass mf
@yacineMTB
Compression is intelligence
Better compression algos are better intelligence
Its pattern recognition + higher "kolmogorov" processing speed
E.g. Waitzkin talks abt his mind going 300fps while his chess opponents mind went 3fps
Wrote abt how I went from being super lazy to getting tons done with less effort+more freedom. one of the most important things in my life
I think this'll help some of you even if you already work hard. Just gotta get through my garbage writing style lol
@justalexoki
Learn by doing (while studying)
Build projects ~15-30% out of your comfort skillset
Each project increases skillset by ~15-30%
Skillset doubles every ~5 projects at 15% and every ~2 at 30% (70/x)
Each proj takes longer but u get faster
get_cracked() in log project complexity
COOOKIIIIIING
IT WOOOORRKSSSS
Invented new network type based on taylor/mclaurin
super fcking based, not even sure if its good or if already discovered (probably tbh this is so basic) but it works
(overfit test btw)
putting energy towards not having to lie leads to waay more success, ime
lying is usually a local optima
put a high cost on low integrity BS and you'll naturally work to find strategies that dont involve lying/deceiving, which are WAY more effective
global optima gang
Did some RL. Set the models fear of death too high
Instead of seeking rewards, it shuffled back/forth to avoid death, getting tons of tiny punishments
Then it learned to die asap to minimize punishments
Reducing its fear fixed it
Over fearing stops success
Reduce it to fix
im gonna say this, and its not gonna make sense, but meta-learning is the single most important capabilities vector ever. literally everything else (math, coding, even agency) is secondary
@Noahpinion
you see what you engage with
curate your feed by both bookmarking and liking a post. Then that kind of content will become disproportionately likely to show up on your FYP
in trough of sorrow, most new ideas look better by comparison
for any large endeavor worth doing, it helps to be aware of this trap
push through the trough. NEVER GIVE UP
ok, as much as I hate this, no more new projects until I finish more old ones
shiny object syndrome is real man
way better to have 4 complete projects than 20 half done ones
wrote a new article:
how you can get tons of useful info
these ideas helped me grow and overcome many huge challenges over the last decade
btw, pls DONT RT unless this has worked for you. if it has, consider RTing. amplify good stuff, filter out slop
@yacineMTB
Use llms but on a much larger scale
For example yesterday for a friend I put 30hrs of huberman podcasts into an llm to answer his question "what does andrew huberman personally do 4hrs before sleep for health" Took like 5 mins (btw just dumping 30hrs of transcripts into a gpt
after im done making my game engine in zig from scratch, i think im gonna pivot to making useful stuff, keep it as free/useful as possible, and add an optional paid plan later or something
even if it doesnt work out, i love building useful tools for myself and they save me time
the lure of freedom is such a crazy good motivator
i had so much fun doing hw in my last yrs of college bc i realized getting way ahead meant i could do whatever i wanted lol
So many ppl on x would love to grind towards a good goal
give em a high payoff long term goal, show em how its possible, and theyll go crazy on it
if I see a pot of gold far away, and i can find a long but totally doable road to it, i start walkin
progress so far on raytraced zig+cuda/opengl game engine from scratch:
- frame image and level geometry stays on the gpu
- camera now works.. 90% correct
hes obsessed
obsessed ppl work like this on autopilot
be obsessed if u wanna work like this. have vision dude. have a hugely ambitious important and possible mission. retire your mom or save the world idk. hes a gaming addict but for his mission
This helped me grow a lot bc I used to try to get things perfect out of fear of mistakes
Literally just do lots and lots of things to grow way faster
Opportunity cost is way worse than mistake cost. Minimize regret gang > minimize mistake cels
the meta game theory is to create a super high trust environment around you
so the local world around you has better games
prisoners dilemma, what? never heard of it lol skill issue
the 5am-9pm 16hr coding session is COMPLETE
next one is monday 6/10! join if you want to do these w a bunch of other ppl studying, building, working, etc. awesome community
Progress
- generated dataset for finetune project
- added database/commands to
16 hr session basically done. actually ended up being 17.5hrs
join our discord, we do these every mon and thurs
huge amount of self improvement actually happening in there. much walking little talking
its like.. avoid the same feeling that makes you want to overpromise. the feeling of wanting to feel like you did something because you told someone you would do it
its painful but so so effective. i now go out of my way to stfu about endeavors im going to do til ive done it
16hr 5-9 work/building session:
- made new ml architectures (mclauren and euler 'polynomial perceptron' models) and made (w LLM lol) an mnist drawing tool for model inference
- made progress on new article (got data w an experiment ive been doing on a method of learning faster)
pro tip for reading ML papers faster/better
watch a yt vid on them first
like this BEAUTIFUL example:
covers residual networks, the context they were created in (imgnet accuracy race), and other stuff you wouldnt learn just from reading the paper 👌👌