Sam Corbett-Davies
@scorbettdavies
Followers
699
Following
1K
Media
12
Statuses
145
Research scientist
Stanford, CA
Joined March 2014
I've successfully defended my PhD! Check out my talk "Identifying bias in human and machine decisions," covering bias in traffic stops, stop-and-frisk, and pretrial algorithms. https://t.co/rUeuTmobOt
7
25
129
Strongly agree. GPT5 writes technical material like it is answering an exam question under time pressure - as if the reader already knows the answer and is grading GPT5. Whereas o3 wrote as if the reader sincerely wanted to know the answer.
I find GPT5 "thinking" to be much less clear than o3 was. Responses are full of jargon, shorthand, much harder to use and interpret. It's like talking to someone who wants to show off his expertise but doesn't really want to help. Is it just me?
0
0
6
Using lung cancer screening as an extended case study, we unpack these four categories of fairness concerns and discuss popular approaches for addressing them. Ultimately, we show that these approaches, if deployed, may in fact WORSEN outcomes for individuals across all groups.
1
1
1
A southern European idea we need to bring to America — bars in playgrounds
60
47
1K
(please retweet) I am taking CS PhD students! If you are interested in ML, data science, health, or inequality, please apply to Cornell :) We have an island campus in New York City and a fantastic group of students I feel lucky to collaborate with every day. Join us!
15
171
608
can I please just have an app to watch fun videos from strangers and an app to see my friends' wedding pics, and not have them all try to be both? please
7
43
1K
Who's attending FAccT in person this year? I'd love to get back to in-person conferences but there's no point if everyone is tuning in remotely!
0
0
1
It’s official, we won the Outstanding Paper Award at #AAAI2022! 🤩 I am so incredibly grateful to my co-authors and mentors @scorbettdavies, Jamal Atif and Nicolas Usunier! @MetaAI @psl_univ @Paris_Dauphine @LAMSADEDauphine @DauphineMiles @RealAAAI
7
11
117
Paper! with @rosemariesays , Bobbie Chern, @scorbettdavies , @mbogen , @tvr2c , @5harad. We propose a strategy to construct more equitable datasets for training ML models. Paper: https://t.co/lz9sfGwXKQ Thread: 1/
2
6
18
The overwhelming evidence for 3rd shot boosters *before* Omicron, reinforced by new data for benefit vs Omicron 1. The randomized trial of 3 shots vs 2 shots + placebo, 95% efficacy for preventing symptomatic infections across all age 16+ groups (N=10,125) https://t.co/UEa1OUlUbk
60
542
2K
Happy to announce our paper “Online certification of preference-based fairness for recommender systems”, accepted to #AAAI2022, w/ @scorbettdavies, J. Atif and Nicolas Usunier! 🎉 @RealAAAI @FacebookAI @DauphineMiles @Paris_Dauphine
https://t.co/5qXMMo5Nuo 1/4
arxiv.org
Recommender systems are facing scrutiny because of their growing impact on the opportunities we have access to. Current audits for fairness are limited to coarse-grained parity assessments at the...
2
5
21
Happy to share our paper “Two-sided fairness in rankings via Lorenz dominance”, accepted to #NeurIPS2021 (w/ @scorbettdavies, J. Atif & Nicolas Usunier) 🎊 We propose a new framework for fair recommendation, grounded in welfare economics. https://t.co/fptTKHJEwD 1/4
arxiv.org
We consider the problem of generating rankings that are fair towards both users and item producers in recommender systems. We address both usual recommendation (e.g., of music or movies) and...
2
4
20
The type of questions that bask in the nuance of fairness and for which there are no mathematically provably right answers. The kinds that make you understand multiple people's perspectives and think about power and who's not being heard.
1
3
39
If you are in California, please take the time to vote in the recall election!!! It's important!
1
4
23
Great thread on how advocates are on course to successfully cripple large swathes of social science research by getting rid of access to anonymized census microdata in order to prevent a hypothetical reconstruction of non-sensitive data that has never been successfully done
/1. Yesterday at the ACS Data Users Conference, the Census Bureau described its plans to replace the American Community Survey (ACS) microdata with “fully synthetic” data over the next three years.
13
121
570
I finally read this @zeynep piece and it is SO good There's a lot that drives me crazy about communication in this pandemic Thing that I find most annoying? The idea that people can't handle nuance Its paternalistic. And untrue Read this https://t.co/HF4H1lU8jF
theatlantic.com
We can learn from our failures.
38
323
1K
Hey all! The @FAccTConference TUTORIAL DEADLINE is FRIDAY (AoE). Submit your amazing tutorials now! https://t.co/9DuHZDFKGg
https://t.co/ovMWA8cRQZ
We are looking for tutorial proposal submissions that address technical, policy, regulatory, ethical, or social science aspects of FAccT issues for a broad audience. They can be translation tutorials or implications tutorials. https://t.co/oYZIyLrH1l
#FAccT2021 Deadline: 18 Dec
0
4
7
Please email your resume to scd at https://t.co/KZ1qBwAKwy if you're interested!
1
2
9
My team at Facebook is hiring quantitative (stats/econ/cs/social science) and qualitative (law/ethics/policy/UXR/humanities/social science) researchers to study fairness across Facebook's products. Intern and full time positions available. Get in touch!
17
63
259
We're hosting a two-day, hands-on data science workshop at @Stanford for Bay Area high school students from traditionally underrepresented groups on Sep 14 & 15! Students receive $200 for participating. Apply by Aug 11, and please help us spread the word!
0
37
42
Are any other US/NZ politics junkies getting Red Peak deja vu from Pete Buttigieg's campaign?
0
0
1