Thank you so much professor. This is really valuable advice. I will definitely follow your tips in writing all of my next papers. Please make the video on how to construct experiments for the research paper when you are free... If I may ask, how much baseline comparisons are sufficient to show that the proposed approach indeed is better than the previous works and at the same time gives some really interesting future directions to explore. Thank you so much professor
In people.cs.umass.edu/~emery/misc/how-to.pdf please read section 2(In this section there is a paragraph starting with "Reading papers is a skill that takes practice..")
Same! I am not able to retain as much information from a paper. As a student, how should we read papers and what do you do when you come across terms/concepts that one is not familiar with?
27:00 Please hate ugly maths as well, but unironically. There are papers that spend lots of time on not so relevant maths in the middle of the paper, not in appendix. "Blending Is All You Need", for example, has gone from integrals and their approximations just to tell go to pseudocode which in python will be "random.choice(vicuna, alpaca).forward(oldcontext+userprompt) leads to better user retention than any model alone". Also ugly graphs that requires color picker to tell lines apart are ugly.
“I encourage my followers to immediately reject any papers with tables they consider ugly”. Please, for the love of everything good, don’t. Give feedback to the authors. There are better things to plant the rejection flag on, than aesthetics or table UX.
Haha, listening to it again, I guess it does come across pretty deadpan. I figured after 30 minutes of deep analysis of the paper, rejecting it because I don't like the table would seem insane even for me.
Great talk! Another video about a timeline for an okay research project would be awesome ;)
Thank you so much professor. This is really valuable advice. I will definitely follow your tips in writing all of my next papers. Please make the video on how to construct experiments for the research paper when you are free... If I may ask, how much baseline comparisons are sufficient to show that the proposed approach indeed is better than the previous works and at the same time gives some really interesting future directions to explore. Thank you so much professor
can we get something on "how to read research papers" too
In people.cs.umass.edu/~emery/misc/how-to.pdf please read section 2(In this section there is a paragraph starting with "Reading papers is a skill that takes practice..")
Same! I am not able to retain as much information from a paper. As a student, how should we read papers and what do you do when you come across terms/concepts that one is not familiar with?
There's a great paper on this, called "How to read a paper" by Keshav. Google it.
Sure... I will add it to the list.
@@srush_nlp Looking forward to this. Thanks
great talk!would watch again and again
27:00 Please hate ugly maths as well, but unironically. There are papers that spend lots of time on not so relevant maths in the middle of the paper, not in appendix. "Blending Is All You Need", for example, has gone from integrals and their approximations just to tell go to pseudocode which in python will be "random.choice(vicuna, alpaca).forward(oldcontext+userprompt) leads to better user retention than any model alone".
Also ugly graphs that requires color picker to tell lines apart are ugly.
25:00 Waiting 🤩
I am wondering where to find the comments "how to make nicer table in LaTex in the comments"
people.inf.ethz.ch/markusp/teaching/guides/guide-tables.pdf
@@srush_nlp Thanks for providing this! :)
“I encourage my followers to immediately reject any papers with tables they consider ugly”.
Please, for the love of everything good, don’t. Give feedback to the authors. There are better things to plant the rejection flag on, than aesthetics or table UX.
Oh come on! That was clearly a joke.
@@srush_nlp I take everything you say seriously 😂
Haha, listening to it again, I guess it does come across pretty deadpan. I figured after 30 minutes of deep analysis of the paper, rejecting it because I don't like the table would seem insane even for me.