@feishaAI

Researcher and engineers in AI and ML; now at Google Research, formerly Professor at USC

 Cambridge, MA
feisha.org
Joined February 2009
  • Tweets713
  • Following341
  • Followers1.9K
30Media
I just watched a great compilation of various people's views about what is coming: xitter.cf/Scr0nkf1nkle/status/19…
Scr0nkf1nkle
@Scr0nkf1nkle29 May 2025
The Great AI Job Displacement Is Closer Than You Think
162
400
58
2,047
406,004
Fei Sha retweeted
ICLR@iclr_conf14 Apr 2025
Announcing the Test of Time awards for ICLR 2025! This award recognizes papers published ten years ago at ICLR 2015 that have had a lasting impact on the field. Congratulations to the authors! blog.iclr.cc/2025/04/14/anno…
4
49
11
506
73,520
LLMs are increasingly used as agents that interact with users. To do so successfully, LLMs need to form beliefs and update them when new information becomes available. Do LLMs do so as expected from an optimal strategy? If not, can we get them to follow this strategy? 🧵
2
76
5
371
47,448
Our team @GoogleAI is hiring an intern. We are interested in having LMs understand and respond to users better. Topics include: teaching LMs to build “mental models” of users; improving LM's reasoning capability over long contexts. @GoogleAI internship deadline is Feb 28.
6
21
1
222
33,010
with @linluqiu @feishaAI @KelseyRAllen @yoonrkim @tallinzen and myself.
1
1
552
I sooo much agree with this. Academic jobs require juggling too many balls with barely any support (grant writing, teaching, managing a research group, supervising students and … research). Take two hrs everyday before you look at email and social media. search.app/nnTKNba5Dqv8EuP48
4
44
7
296
33,614
Don't miss the Future of Machine Learning Symposium on July 19, 2024 at @ISTAustria! Join leading experts to explore advancements and emerging trends in #MachineLearning. Register now 🔗 fml2024.ista.ac.at/ @feishaAI @KaterinaFragiad @tkipf @mmbronstein @riken_en @MIT
6
18
4,210
Are you planning to attend #ICML2024 in Vienna? Why not come two days earlier and attend the Symposium "The Future of Machine Learning" at @ISTAustria on July 19th? We'll have great speakers, free coffee, and enough space to meet and chat. fml2024.ista.ac.at @icmlconf
2
14
2
47
8,659
On Tuesday, 10:12 am at "Don Alberto 1" at #naacl2024 @tiwa_eisape will present his Google student researcher project where we compare humans' and LLMs' reasoning fallacies! (with @vansteenkiste_s Dasgupta @feishaAI @mhtessler) arxiv.org/abs/2311.00445

A Systematic Comparison of Syllogistic Reasoning in Humans and...

A central component of rational behavior is logical inference: the process of determining which conclusions follow from a set of premises. Psychologists have documented several ways in which...

arxiv.org
1
4
1
35
3,819
If you're a PhD student interested in interning with me or one of my amazing colleagues at Microsoft Research New England (@MSRNE, @MSFTResearch) this summer, please apply here jobs.careers.microsoft.com/g…
11
79
3
392
86,513
Do deep transformer LMs generalize better? In a new preprint we (@vansteenkiste_s Ishita Dasgupta @feishaAI @dhgarrette @tallinzen) control for parameter count to show how depth helps models on compositional generalization tasks, but diminishingly so🧵 📜 jpetty.org/depth
2
24
5
80
24,499
We are excited to announce new work developing diffusion models for predictions and uncertainty estimation for chaotic dynamical systems and rare events. Appearing at #ICML2023 today! With @m_finzi, @bloopsie, @feishaAI, Leonardo Zepeda-Núñez. arxiv.org/abs/2306.07526
1
11
78
8,257
New paper! Retrieval-augmented models are expensive. Make them faster by partially pre-computing passage representations. Work with @_theopompus, @nfitz, Joshua Ainslie, Sumit Sanghai, @feishaAI, @professorwcohen. 1/N arxiv.org/abs/2301.10448

Pre-computed memory or on-the-fly encoding? A hybrid approach to...

Retrieval-augmented language models such as Fusion-in-Decoder are powerful, setting the state of the art on a variety of knowledge-intensive tasks. However, they are also expensive, due to the...

arxiv.org
1
14
2
100
15,787
People are testing large language models (LLMs) on their "cognitive" abilities - theory of mind, causality, syllogistic reasoning, etc. Many (most?) of these evaluations are deeply flawed. To evaluate LLMs effectively, we need some principles from experimental psychology.🧵
36
460
78
1,822
549,143
Fei Sha retweeted
Google AI
@GoogleAI31 Mar 2023
Learn about ViT-22B, the result of our latest work on scaling vision transformers to create the largest dense vision model. With improvements to both the stability and efficiency of training, ViT-22B advances the state of the art on many vision tasks → ai.googleblog.com/2023/03/sc…
33
124
11
520
188,577
Distinguished Speaker📚📚: "Thank you, Dr. @feishaAI! Your talk was thought-provoking & challenged us to think about the future of #AI & the next generation of LLMs & generative AI technologies,” @Bestavros bit.ly/3G2T3sM
BU Faculty of Computing & Data Sciences

ALT BU Faculty of Computing & Data Sciences

1
1
1
12
1,683
Terrifyingly hilarious overview of an insane number of mistakes in last week’s Bing/ChatGPT demo. Why did Google lose 10% of their value for a technicality, but Microsoft threw up 50 minutes of bullshit and no one noticed? xitter.cf/dkbrereton/status/1625…
dmitri@dkbrereton13 Feb 2023
The people who are hyped about Bing AI must have seen a different demo than I did. It made several mistakes during the demo, much worse than Google's Bard mistake. Here's what Bing AI got wrong:
35
62
8
564
234,887
.@ChrisGPotts has already provided some references supporting alternative views from @spiantado, @FelixHill84, @lambdaviking & @tallinzen, but he was too modest to mention his own blog post on the topic, which I rather like: 5/ chrisgpotts.medium.com/is-it…

Is it possible for language models to achieve language understanding?

I was invited to deliver a few remarks at an HAI symposium on OpenAI’s GPT-3 project, at the end of October. I chose for my title “Is it…

chrisgpotts.medium.com
3
6
2
103
A few more hours left to register for what may be the most exciting PhD program in Europe
ELLIS@ELLISforEurope5 Oct 2022
Looking for PhD opportunities in #MachineLearning or related fields? The #ELLISPhD Program is now accepting applications! Apply until November 15, 2022. #JoinELLISforEurope #phdlife #phd #AI ellis.eu/news/ellis-phd-prog…

ELLIS PhD Program: Call for applications 2022

The ELLIS PhD program is a key pillar of the ELLIS initiative whose goal is to foster and educate the best talent in machine learning and related research areas by pairing outstanding students with...

ellis.eu
1
13
57
A Transformer with billions of parameters can learn a lot from images and text on the web A small MLP with 100 parameters, not so much
Gary Marcus
@GaryMarcus30 May 2022
A human can learn a lot from college, a parrot, not so much.
2
4
16
Load more