Vihari (@p_vihari) / X

Vihari

150 posts

Vihari

@p_vihari

Research Scientist @GoogleDeepMind Past: Postdoc @CambridgeMLG @ChurchillCol PhD @CSE_IITBombay

Bengaluru

Joined February 2012

Vihari
@p_vihari
Jan 5, 2024
To all the prospective PhDs in/from India, before you sign up for a PhD position in the US, please carefully consider if you would do better (PhDing) in India. I sketched my experience in this blog vihari.github.io/why_india/. Feel free to send me your comments 😃
vihari.github.io
Vihari Piratla
Grass is just as green on our side! **TL; DR**. I refused to enrol as a PhD student in India for many years because I was ill-informed, but I finally did in 2019 and had a great experience. This...
85K
Vihari
@p_vihari
Nov 1, 2022
Replying to @GabbbarSingh and @St_Hill
Well! Trice, Trite, Trave, Trier, Trace, Track, Treat, Triac, Tract, Trica, Tribe, Traik, Trick, Trike, or Trait.
Vihari
@p_vihari
Dec 14, 2023
If you are at NeurIPS, please visit our poster #1618 between 10:45-12:45 today. Learning from explanations (and other forms of richer task specification) is my bet for the future of reliable systems. Drop by to know more 😃 with @juyeonheo2324, @adrian_weller @matthew_wicker
1.5K
Vihari
@p_vihari
Apr 24, 2022
I am excited to be presenting our recent work on sub-population shift problem this Tuesday at #ICLR2022 Hope to see you all there. 📢Poster: iclr.cc/virtual/2022/p… Paper: arxiv.org/abs/2110.02619 Slides: docs.google.com/presentation/d… @PNetrapalli @aimliitbcse
Vihari
@p_vihari
Feb 19, 2024
I wrote a short summary of AI Alignment that you may find handy given the many (somewhat needlessly introduced) terms. I also included some personal thoughts. Happy to hear your thoughts.
Short Summary of AI Alignment
From gist.github.com
1.2K
Vihari
@p_vihari
Aug 19, 2021
Delighted to be presenting at TrustML Seminar. I will be talking about research challenges when serving millions of clients through Machine Learning Service APIs. Thank you @trustworthy_ml for having me. The event starts at 9:30 PM IST, 12PM ET. Please do join.
Trustworthy ML Initiative (TrustML)
@trustworthy_ml
Aug 17, 2021
1/ It’s Rising Star Spotlights Seminar ⭐️ time again! For this week’s TrustML seminar, we're delighted to host @shiorisagawa (Stanford) & @p_vihari (IITB) on Thurs Aug 19th 12pm ET 🎉🎉🥳 Register here: us02web.zoom.us/webinar/regist… See this thread for the speaker & talk details👇
Vihari
@p_vihari
May 10, 2023
Replying to @mishumausam and @aclmeeting
I vote to keep them all. 1. Unlike stacked transformers etc., we understand PCFG, FSAs better. Satisfaction factor. 2. Classical methods are bridges to understand workings and intent of modern methods. 3. Know when to use simpler tools.
663
Vihari
@p_vihari
Apr 27, 2023
OK, here we go. A course on "how to dance to the whims of LLMs". Also, why call it engineering?
Andrew Ng
@AndrewYNg
Apr 27, 2023
1/ Thrilled to announce: Our new course ChatGPT Prompt Engineering for Developers, created together with @OpenAI, is available now for free! Access it here: deeplearning.ai/short-courses/…
00:00
249
Vihari
@p_vihari
Dec 29, 2023
Come to think of it, LLMs are a raging hit partly because they are never pitched to be perfect or personified (unlike other AI tools in the past). Any newly found shortcoming of LLMs is more a caution to the user than an issue to its makers. Brilliant!
1.2K
Vihari
@p_vihari
Jan 24, 2024
Replying to @iKDD_News and @iitbombay
Very grateful for the recognition. Massive thanks to the organizers and award committee for all their efforts. For anyone interested, here is the link to my thesis arxiv.org/abs/2303.02781
168
Vihari
@p_vihari
Dec 16, 2023
If you are at NeurIPS, please visit our poster(s) at XAIA workshop. Concept explanations are my favorite kind, and we proposed a low-variance estimator that enabled estimation even with about thousand concepts with @juyeonheo2324, S. Singh, @adrian_weller
Cambridge MLG
@CambridgeMLG
Dec 15, 2023
Replying to @CambridgeMLG
And finally, find any of the other papers/talks at the other workshops listed below!
arxiv.org
Estimation of Concept Explanations Should be Uncertainty Aware
Model explanations can be valuable for interpreting and debugging predictive models. We study a specific kind called Concept Explanations, where the goal is to interpret a model using...
1.3K
Vihari
@p_vihari
Jul 10, 2023
Grand scheme of things. More hype about super-human AI => more funding/budget/race to build one. "Rockstar" robot (of the Jam Band) needs a human to hold the mic 😂
Reuters
@Reuters
Jul 7, 2023
Nine humanoid robots gathered at the United Nations’ 'AI for Good' conference in Geneva for the world’s first human-robot press conference reut.rs/3PNvNor
00:00
252
Vihari
@p_vihari
Jan 5, 2024
Yes, I can relate with that. Having to write a competitive exam sucks. But may only require a 2-3 months of preparation if you are a strong candidate (I am just speculating, I have no idea how hard the exam really is).
1.1K
Vihari
@p_vihari
Jan 5, 2024
Replying to @yashgyy
Sure, there are exceptions. I still think the path is relatively harder.
2K