Hi! In 2020 I found your channel and have been keeping up with your work! You've been doing amazing stuff! That being said, I never managed to get a "Better Poster" into a poster presentation. I always tried to give it as an option, but my supervisors would essentially shut it down. The problem, from their perspective, was that the better poster is essentially too nebulous. For what I was doing, the amount on information that needed to be written just does not fit into a better poster. Having less information does not work as well - as you need the information to have the conclusions. This poster design would work in a big conference of random scientists, but not in a singular topic conference. In a single topic conference you really need to give all the details. That being said, I am very interested in the future of better science, because some posters are just the worst.
Thank you so much! Getting betterposters past supervisors is a common struggle. Couple questions: 1. Really, the takeaways and big figures matter more than the particular layouts. Is that doable? 2. How many posters in a room at the specialist conferences? 3. Are presenters standing next to them?
8:34 Do you mean that we use tools like StableDiffusion to generate illustrations to get our points across? I had thought about the possibility of generating full presentations, but getting good training data in a format that is easy to process (i.e. not ugly beamer presentations, even though they are very easy to process if you have the code) seems quite difficult.
Yep exactly! Just stablediffusion images instead of stock photos. Long term though I hope generative images can become good enough that you can accurately visualize things that we don't have the power to observe dirtectly. Full presentation visualization would be wonderful if it was good, but yeah how do you get training data when your job is to invent brand-new knowledge?
What about research that transforms the way research is done or should be done? I’m an AI researcher and even there not everything is amenable to modular method drop-in. For example, back in the day HIV was extremely marginalized due to the language of the research and discussions of the people, and improving that changed the research directions and outcomes. An AI extracted snippet can’t capture that. Btw I refer to your better posters video (#2) all the time, short summaries are definitely useful, but not always sufficient!
Your HIV example is beautiful (in a tragic way, ofc) and makes a good point. I'm still kind of wondering about the place for human narrative ability in the AI scientific-article future. I guess one question: Do you think there will always be an article? Or do you think we'll eventually consume all content through AI summary of data/studies? Asking generally and for science I guess. BTW - The endgame is still progressive disclosure. As much information as you want, when you want it. Without ever overloading you, or hitting bottom. It's way, way easier to accomplish that with articles than posters though! (especially computational articles)
Hi! In 2020 I found your channel and have been keeping up with your work! You've been doing amazing stuff!
That being said, I never managed to get a "Better Poster" into a poster presentation. I always tried to give it as an option, but my supervisors would essentially shut it down. The problem, from their perspective, was that the better poster is essentially too nebulous. For what I was doing, the amount on information that needed to be written just does not fit into a better poster. Having less information does not work as well - as you need the information to have the conclusions. This poster design would work in a big conference of random scientists, but not in a singular topic conference. In a single topic conference you really need to give all the details.
That being said, I am very interested in the future of better science, because some posters are just the worst.
Thank you so much! Getting betterposters past supervisors is a common struggle. Couple questions:
1. Really, the takeaways and big figures matter more than the particular layouts. Is that doable?
2. How many posters in a room at the specialist conferences?
3. Are presenters standing next to them?
8:34 Do you mean that we use tools like StableDiffusion to generate illustrations to get our points across? I had thought about the possibility of generating full presentations, but getting good training data in a format that is easy to process (i.e. not ugly beamer presentations, even though they are very easy to process if you have the code) seems quite difficult.
Yep exactly! Just stablediffusion images instead of stock photos. Long term though I hope generative images can become good enough that you can accurately visualize things that we don't have the power to observe dirtectly. Full presentation visualization would be wonderful if it was good, but yeah how do you get training data when your job is to invent brand-new knowledge?
What about research that transforms the way research is done or should be done?
I’m an AI researcher and even there not everything is amenable to modular method drop-in.
For example, back in the day HIV was extremely marginalized due to the language of the research and discussions of the people, and improving that changed the research directions and outcomes. An AI extracted snippet can’t capture that.
Btw I refer to your better posters video (#2) all the time, short summaries are definitely useful, but not always sufficient!
Your HIV example is beautiful (in a tragic way, ofc) and makes a good point. I'm still kind of wondering about the place for human narrative ability in the AI scientific-article future. I guess one question: Do you think there will always be an article? Or do you think we'll eventually consume all content through AI summary of data/studies?
Asking generally and for science I guess.
BTW - The endgame is still progressive disclosure. As much information as you want, when you want it. Without ever overloading you, or hitting bottom.
It's way, way easier to accomplish that with articles than posters though! (especially computational articles)
🚀💚