Neeraj Varshney

Applied Scientist at Amazon | PhD @ ASU | BITS Pilani

[Neeraj Varshney @nrj_varshney]

Whether you're a researcher, engineer, or just curious about how modern LLMs evolve beyond pretraining — this is for you.

Would love to hear your thoughts and feedback!

#LLMs #AI #PostTraining #NLP #MachineLearning #RLHF #ConstitutionalAI #DPO #SFT

[Neeraj Varshney @nrj_varshney]

I'm preparing a detailed article series on Post-Training of Large Language Models (LLMs) – the techniques that make models like ChatGPT truly useful after pretraining.

🧠 The first part is here https://t.co/tTgVZ2UCuV

[Neeraj Varshney @nrj_varshney]

@keviv9 @SCAI_ASU @ASU @CityofPhoenixAZ Congrats Vivek!

[Neeraj Varshney @nrj_varshney]

@paul_rottger Thanks Paul.
This is a great list.
Sharing our recent work https://t.co/xxSr1f3yd0
in which we compare various LLM defense strategies on both 'safety' and 'over-defensiveness'. Hope you will find it interesting.

[Neeraj Varshney @nrj_varshney]

@chipro Sharing my work pertinent to #1 Hallucinations of LLMs.

The proposed approach actively detects and mitigates hallucinations during the generation process and also facilitates in preventing their propagation in the model's output.
https://t.co/SCnYVPI5Y9

[Neeraj Varshney @nrj_varshney]

I'm thrilled to share that my 2 papers got accepted to #ACL2023
One on "improving reliability of QA systems" & the other on "addressing novelties in NLP task"

Also, this semester, I have completed my PhD compre exam and thesis proposal and have now advanced to "PhD candidacy"

[Neeraj Varshney @nrj_varshney]

I am excited to share that my paper "Can Open-Domain QA Reader Make Efficient Use of External Knowledge like Humans?" has been accepted to appear at the AAAI'23 Workshop on Knowledge Augmented Methods for NLP.
#AAAI23 #KnowledgeNLP #NLProc

[Neeraj Varshney @nrj_varshney]

I published a fun Trivia app on @huggingface spaces.
Give it a topic and it will ask you an interesting question.
Check it out here:
https://t.co/N4Y79BRuTy

[Micah Goldblum @micahgoldblum]

A common point raised by ML reviewers is that a method is too simple or is made of existing parts. But simplicity is a strength, not a weakness. People are much more likely to adopt simple methods, and simple ones are also typically more interpretable and intuitive. 1/2

[Neeraj Varshney @nrj_varshney]

In-person conferences are back finally!
#ACL2022 #NLProc @aclmeeting
Ecstatic to present my papers and meet researchers from all over the world under a single roof.
Here is the schedule of my poster sessions.

[Ai2 @allen_ai]

Greetings all #ACL2022 attendees! AI2 has a lot of great work we're representing at this year's conference; check out our roundup of AI2 talks, workshops, tutorials, and papers here to learn where to find us:
https://t.co/wLcJHaTeMn

[Yizhong Wang @yizhongwyz]

Is it possible to solve NLP tasks by simply following instructions that define the tasks? How can we measure the progress?

Excited to announce Natural Instructions v2, a collection of 1600+ diverse language tasks and their expert-written instructions!

📜https://t.co/DBsz7KSBXt

[Neeraj Varshney @nrj_varshney]

3. Investigating Selective Prediction Approaches Across Several Tasks in IID, OOD, and Adversarial Settings
4. NumGLUE: A Suite of Fundamental yet Challenging Mathematical Reasoning Tasks
#acl2022

[Neeraj Varshney @nrj_varshney]

I am happy to share that my 4 papers got accepted to ACL 2022
@aclmeeting
#acl2022
1. ILDAE: Instance-Level Difficulty Analysis of Evaluation Data
2. Unsupervised Natural Language Inference Using PHL Triplet Generation
...

[Neeraj Varshney @nrj_varshney]

A Primer on Knowledge Distillation in NLP — Part 1 https://t.co/N0nrLIj46A

[Ari Holtzman @universeinanegg]

"You can't learn language from the radio." 📻

Why does NLP keep trying to?

In https://t.co/yWEnq5QW9R we argue that physical and social grounding are key because, no matter the architecture, text-only learning doesn't have access to what language is *about* and what it *does*.

[Dan Jurafsky @jurafsky]

Here's the 2020 release of draft chapters for Speech and Language Processing, just in time to wish you all a Happy New Year! Enjoy! https://t.co/pjfZsYfJ5S

[Neeraj Varshney @nrj_varshney]

Check out my article on Out-of-Distribution Detection in Deep Neural Networks https://t.co/Htws2Sg3eu

[Neeraj Varshney @nrj_varshney]

Excited to share my article series on Multi-task learning.
A Primer on Multi-task Learning — Part 1 https://t.co/JHfnahcwtp

A Primer on Multi-task Learning — Part 2
https://t.co/QFuSngHjio

[Greg Brockman @gdb]

OpenAI for converting bullet points into a proper email, now as a Gmail plugin: