Natalie Collina

58 posts

Natalie Collina

Natalie Collina

@natalie_collina

CS PhD student @Penn studying strategic human-AI interaction. @Google, @Princeton.

Philadelphia, PA Katılım Mayıs 2024
250 Takip Edilen461 Takipçiler
Natalie Collina
Natalie Collina@natalie_collina·
Probably not gonna hang around here for much longer. Find me on 🦋:)
English
1
0
6
1.1K
Natalie Collina retweetledi
EnCORE Institute
EnCORE Institute@EnCOREInstitut·
EnCORE student seminar is an opportunity to get a sneak peek of our fabulous talent pool and learn about their work. Join us on Oct 28th as Chhavi and Natalie tell us how AI models can be fair proof and give better predictions. @chhaviyadav_ @natalie_collina @Aaroth @kamalikac
EnCORE Institute tweet media
English
0
5
14
6.8K
Natalie Collina retweetledi
Aaron Roth
Aaron Roth@Aaroth·
Calibration in 15 minutes: youtu.be/ZXkjZ8f0sJw (This is my 15 minute tutorial on calibration that I'll give to kick of the FOCS calibration workshop this Sunday. Not too late to buy your plane ticket to Chicago!)
YouTube video
YouTube
Aaron Roth@Aaroth

In just 12(!) days we'll be hosting this neat workshop on calibration at FOCS: ideal-institute.org/2024/09/10/foc… There is a great lineup of speakers including Parikshit Gopalan, @natalie_collina, Michael Kim, @adamfungi and Princewell Okoroafor. There is still time to book your trip!

English
0
9
37
5.1K
Natalie Collina
Natalie Collina@natalie_collina·
Forgot to post this until I already gave my talk, but I’m in Seattle for INFORMS this year! If you’re also here and interested in online learning/game theory/human-AI interaction, let me know—I truly can’t figure out how to use the app 😅
English
1
0
13
925
Natalie Collina
Natalie Collina@natalie_collina·
Super excited to speak alongside this amazing group of calibration researchers! I’ll be discussing our new research on designing protocols so that multiple calibrated decision makers can reach agreement and improve accuracy.
Aaron Roth@Aaroth

In just 12(!) days we'll be hosting this neat workshop on calibration at FOCS: ideal-institute.org/2024/09/10/foc… There is a great lineup of speakers including Parikshit Gopalan, @natalie_collina, Michael Kim, @adamfungi and Princewell Okoroafor. There is still time to book your trip!

English
0
0
17
1.7K
Natalie Collina
Natalie Collina@natalie_collina·
I’m interested in learning more about LLMs in games (e.g. chess, their ability to construct and respond to strategic behavior generally) and games in LLMs (e.g. game-theoretic ideas in RLHF). If anyone has any recommendations for papers to read, please let me know!
English
19
13
256
42.6K
Aaron Roth
Aaron Roth@Aaroth·
Its also important to be unafraid to say stupid things (and to foster an environment where other are also unafraid). Most first thoughts are stupid, but it is important to say them so that you can get to second and third thoughts.
English
3
4
59
2.9K
Aaron Roth
Aaron Roth@Aaroth·
It is under-appreciated that -thinking- is a social activity. The best ideas I have had are in conversation. This is one of the most important reasons to schedule regular research meetings, even when it seems like there aren't lots of new developments to talk about.
English
1
20
164
9.4K
Natalie Collina retweetledi
Itai Yanai
Itai Yanai@ItaiYanai·
Doing good science is 90% finding a science buddy to constantly talk to about the project.
Itai Yanai tweet media
English
177
3.2K
17.8K
1.8M
Jarushka Naidoo
Jarushka Naidoo@DrJNaidoo·
Superb article on the secret sauce of team science, in @NatureBiotech: It takes two to think ‘Morale thrives on the encouragement we get from the feedback of someone who understands our work & it’s motivations’ Great read nature.com/articles/s4158…
Jarushka Naidoo tweet media
English
6
62
223
41.7K
Gokul Swamy
Gokul Swamy@g_k_swamy·
@natalie_collina @EugeneVinitsky Yay! You might also find this follow-up where we detail how best to instantiate the reduction in LLM-land pretty interesting as well: arxiv.org/abs/2404.16767. There’s some really beautiful connections to online mirror descent / natural policy gradient included as a treat :)
English
1
0
1
108