Also, I am planning on doing a series of how-to videos, mostly focusing on taking limited amount of concepts from technical reports and academical papers and show people how to find the corresponding code from those concepts. It will be about approaching research with an engineering mindset, please let me know if you guys will be interested in it. Will drop one on how-to read deepseek-v3 tomorrow, remember to check it out guys!
Definitely lol. I’m taking a grad research course on some more advanced ML topics and while I’m figuring out for myself how to actually approach topics in research, I wanna hear your perspective. I’ve taken a fair amount of math and stats courses before, but yea def a challenge haha. I was talking to one of my other undergrad friends in the class and we r all mad confused haha, and he’s taken way more advanced classes than me.
Just to let anyone following this to know. I have just finished recording all the content but it need some reorganization and a little bit editing. Check it out in a day. :)
Thanks for making this video! The analogy I thought of while watching is abstraction layers. When there’s a passed-down conclusion like “why we do early-stopping”, or “how cell Meiosis works”, it’s generally a very useful abstraction that helps us avoid a lot of unnecessary thinking. But if we want to work in a field, that thinking becomes necessary again! We’ll want to re-investigate accepted conclusions, just in case: - the conclusion is an imperfect/simplified analogy - the field has changed, so previous conclusions are no longer valid (such as the early-stopping example) - someone lied (rare in AI papers, common in some other fields!) Plus, re-creating previous results helps us get practice in the tools and ways of thinking needed to do original research. I also really liked the why vs why-not distinction you made - believing by default vs disbelieving by default. From my read of the Musk biography, lots of his success come from simply *not believing people* when they say something has to be done a certain way.
The idea is to not to take other people's word for something without verifying it at some practical level. I think this idea does resonate with the first principle which Musk mentions a lot.
@@ezmonyi Great work! Main response in a top-level thread, but burying my comments on video-making in a thread so it's not what people see first :) 1. The #1 production-quality improvement would be a better mic. You can do an entire channel without a camera or fancy editing, but it's stressful to listen to low audio quality. You can get a high-quality mic for $100 - my first was a Blue Yeti and it was great. 2. The #2 production-quality improvement would be uploading in a higher resolution Also, could you link to your comment when replying to this, so I can re-pin it? After you edited to add the link I can no longer find it, I think my suggestion to edit the link in was not a good one based on how youtube handles comments :(
you know what bro ; i was and now not like in the past a very hardworker in math , by doing that i integrated a very good university where i meet guys that are briallants and had a tallent , so due to my social intelligence , i create the contact with them and they've became my friends ; what i found is that there are smart , but unidimentional ; and they worked too much and even to the point of dissapearing , so here i found the key , i need to stay with one of them and see how he works , what i found is : no distraction , no phone at that time , also just math, solving problems and sleeping! here i found that it's not enough to just read a book or watch a course in the hell loop of courses , and me too until now i have that kind of problem ; like oh the time is going to fast i need to do something ? what is the ' something i need to do ' i don't know ; so the mean thing is to make a goal onto eyes and try to ask and create new stuff by practicing , learn by doing
Not completely sure of your point, but I am also a guy who talks out of my mind, so I can get your vibe. I would say my approach on life is that almost everything that propel you to more action and more real life feedback is good for me, at least when I am still young and ignorant on the way the world works. Do more, expect greatness, see the gap between your expectation and reality, adjust your heuristics, things will be great brother.
So, in summary what I'm taking away is: Always inquire a bit deeper into any conclusive statements (from external resources) as the statements are not as useful to beginners until they are applied / illustrated / implemented / played with in some form and in the current context of DL/ML/AI (mainly code or experiments)...is that right?
Yes. In DL/ML/AI experiment is always much more insightful than theory/conclusive statements. It's almost never what you think it is and there are so many variable in practice. Limiting yourself on a level of verbal reason or mathematical theory is just confining it to a limited past understanding. Talk less, do more.
Also, I am planning on doing a series of how-to videos, mostly focusing on taking limited amount of concepts from technical reports and academical papers and show people how to find the corresponding code from those concepts. It will be about approaching research with an engineering mindset, please let me know if you guys will be interested in it. Will drop one on how-to read deepseek-v3 tomorrow, remember to check it out guys!
Definitely lol. I’m taking a grad research course on some more advanced ML topics and while I’m figuring out for myself how to actually approach topics in research, I wanna hear your perspective. I’ve taken a fair amount of math and stats courses before, but yea def a challenge haha. I was talking to one of my other undergrad friends in the class and we r all mad confused haha, and he’s taken way more advanced classes than me.
Just to let anyone following this to know. I have just finished recording all the content but it need some reorganization and a little bit editing. Check it out in a day. :)
Thanks for making this video!
The analogy I thought of while watching is abstraction layers. When there’s a passed-down conclusion like “why we do early-stopping”, or “how cell Meiosis works”, it’s generally a very useful abstraction that helps us avoid a lot of unnecessary thinking.
But if we want to work in a field, that thinking becomes necessary again! We’ll want to re-investigate accepted conclusions, just in case:
- the conclusion is an imperfect/simplified analogy
- the field has changed, so previous conclusions are no longer valid (such as the early-stopping example)
- someone lied (rare in AI papers, common in some other fields!)
Plus, re-creating previous results helps us get practice in the tools and ways of thinking needed to do original research.
I also really liked the why vs why-not distinction you made - believing by default vs disbelieving by default. From my read of the Musk biography, lots of his success come from simply *not believing people* when they say something has to be done a certain way.
The idea is to not to take other people's word for something without verifying it at some practical level. I think this idea does resonate with the first principle which Musk mentions a lot.
Thanks! I am stuck in this loop. Gonna try and escape it and work on some practical projects.
Heuristics + Bayes + some Ad Hocs.
Thanks! That was eye opening! I would love to see more videos about practical examples
Thanks! Also if there's any suggestion on video making, feel free to drop them, I am the newbie in that.
@@ezmonyi Great work! Main response in a top-level thread, but burying my comments on video-making in a thread so it's not what people see first :)
1. The #1 production-quality improvement would be a better mic. You can do an entire channel without a camera or fancy editing, but it's stressful to listen to low audio quality. You can get a high-quality mic for $100 - my first was a Blue Yeti and it was great.
2. The #2 production-quality improvement would be uploading in a higher resolution
Also, could you link to your comment when replying to this, so I can re-pin it? After you edited to add the link I can no longer find it, I think my suggestion to edit the link in was not a good one based on how youtube handles comments :(
you know what bro ; i was and now not like in the past a very hardworker in math , by doing that i integrated a very good university where i meet guys that are briallants and had a tallent , so due to my social intelligence , i create the contact with them and they've became my friends ; what i found is that there are smart , but unidimentional ; and they worked too much and even to the point of dissapearing , so here i found the key , i need to stay with one of them and see how he works , what i found is : no distraction , no phone at that time , also just math, solving problems and sleeping! here i found that it's not enough to just read a book or watch a course in the hell loop of courses , and me too until now i have that kind of problem ; like oh the time is going to fast i need to do something ? what is the ' something i need to do ' i don't know ; so the mean thing is to make a goal onto eyes and try to ask and create new stuff by practicing , learn by doing
Not completely sure of your point, but I am also a guy who talks out of my mind, so I can get your vibe. I would say my approach on life is that almost everything that propel you to more action and more real life feedback is good for me, at least when I am still young and ignorant on the way the world works. Do more, expect greatness, see the gap between your expectation and reality, adjust your heuristics, things will be great brother.
So, in summary what I'm taking away is:
Always inquire a bit deeper into any conclusive statements (from external resources) as the statements are not as useful to beginners until they are applied / illustrated / implemented / played with in some form and in the current context of DL/ML/AI (mainly code or experiments)...is that right?
Yes. In DL/ML/AI experiment is always much more insightful than theory/conclusive statements. It's almost never what you think it is and there are so many variable in practice. Limiting yourself on a level of verbal reason or mathematical theory is just confining it to a limited past understanding. Talk less, do more.