Desh Raj
banner
rdesh26.bsky.social
Desh Raj
@rdesh26.bsky.social
Research Scientist @ Meta GenAI in NYC.
Working on audio/speech for LLaMA.

Previously: PhD @ JHU CLSP

desh2608.github.io
Reposted by Desh Raj
Extremely excited to announce that I will be joining
@utaustin.bsky.social Computer Science in August 2025 as an Assistant Professor! 🎉
May 5, 2025 at 8:28 PM
Reposted by Desh Raj
Johns Hopkins is the largest private employer in Maryland.
March 13, 2025 at 9:48 PM
Reposted by Desh Raj
I got laid off today, with the rest of 18F.

18F was an elite federal software shop. We made gov't websites work better, more efficiently for the American people. We saved taxpayers from getting screwed over by contractors. And were fired for it.

We made this website to tell our story:
18f.org
We're not done yet | 18F
18f.org
March 1, 2025 at 10:38 PM
Reposted by Desh Raj
For decades, the US government has painstakingly kept American science #1 globally—and every facet of American life has improved because of it. The internet? Flu shot? Ozempic? All grew out of federally-funded research. Now all that's being dismantled. 1/ www.technologyreview.com/2025/02/21/1...
The foundations of America’s prosperity are being dismantled
Federal scientists warn that Americans could feel the effects of the new administration's devastating cuts for decades to come
www.technologyreview.com
February 21, 2025 at 1:01 PM
Reposted by Desh Raj
Insightful post by Kyunghyun Cho on the job prospects of PhD students finishing right now. AI has seen so rapid change that the world isn't the same it used to be when they started their studies.

kyunghyuncho.me/i-sensed-anx...
i sensed anxiety and frustration at NeurIPS’24 – Kyunghyun Cho
kyunghyuncho.me
December 24, 2024 at 9:44 AM
Reposted by Desh Raj
Entropy is one of those formulas that many of us learn, swallow whole, and even use regularly without really understanding.

(E.g., where does that “log” come from? Are there other possible formulas?)

Yet there's an intuitive & almost inevitable way to arrive at this expression.
December 9, 2024 at 10:44 PM
Reposted by Desh Raj
Inventors of flow matching have released a comprehensive guide going over the math & code of flow matching!

Also covers variants like non-Euclidean & discrete flow matching.

A PyTorch library is also released with this guide!

This looks like a very good read! 🔥

arxiv: arxiv.org/abs/2412.06264
December 10, 2024 at 8:35 AM
Reposted by Desh Raj
🚨I too am on the job market‼️🤯

I'm searching for faculty positions/postdocs in multilingual/multicultural NLP, vision+language models, and eval for genAI!

I'll be at #NeurIPS2024 presenting our work on meta-evaluation for text-to-image faithfulness! Let's chat there!

Papers in🧵, see more: saxon.me
December 6, 2024 at 1:44 AM
Reposted by Desh Raj
🚨 I am on the faculty job market this year 🚨
I will be presenting at #NeurIPS2024 and am happy to chat in-person or digitally!

I work on developing AI agents that can collaborate and communicate robustly with us and each other.

More at: esteng.github.io and in thread below

🧵👇
December 5, 2024 at 7:00 PM
Reposted by Desh Raj
This is my first official post at Bluesky with great news :)

We got the best paper award at IEEE SLT'24! This work elegantly and straightforwardly solves contextual biasing issues with dynamic vocabulary arxiv.org/abs/2405.13344. Congrats, Yui, Yosuke, Shakeel, and Yifan!
! I'm super happy!
December 4, 2024 at 2:16 PM
Reposted by Desh Raj
I am seriously behind uploading Learning Machines videos, but I did want to get @jonathanberant.bsky.social's out sooner than later. It's not only a great talk, it also gives a remarkably broad overview and contextualization, so it's an excellent way to ramp up on post-training
youtu.be/2AthqCX3h8U
Jonathan Berant (Tel Aviv University / Google) / Towards Robust Language Model Post-training
YouTube video by Yoav Artzi
youtu.be
December 2, 2024 at 3:45 AM
Reposted by Desh Raj
I just learned that Torch ctc_loss calculates the wrong gradient (but when there was log_softmax before, it does not matter).

For the grad ctc_loss w.r.t. log_probs, it calculates exp(log_probs) - y, but correct would be -y. Some workaround: github.com/pytorch/pyto...

PS: First Bluesky post.
CTCLoss gradient is incorrect · Issue #52241 · pytorch/pytorch
🐛 Bug Hi, While working on some CTC extensions, I noticed that torch's CTCLoss was computing incorrect gradient. At least when using CPU (I have not tested on GPU yet). I observed this problem on b...
github.com
November 26, 2024 at 11:16 PM
Can someone recommend good resources for deep dive into RLHF methods? I have a basic understanding mostly from blogs and mainstream papers (like InstructGPT), but would like to gain more insights + understand current research. Thanks in advance!
November 27, 2024 at 2:51 AM
Reposted by Desh Raj
I've started putting together a starter pack with people working on Speech Technology and Speech Science: go.bsky.app/BQ7mbkA

(Self-)nominations welcome!
November 19, 2024 at 11:13 AM
Reposted by Desh Raj
Does everyone in your community agree on some folk knowledge that isn’t published anywhere? Put it in a paper! It’s a pretty valuable contribution
November 26, 2024 at 10:31 PM
Our GenAI-Speech team at Meta is hiring RS interns for summer 2025 to work on speech, LLMs, dialog generation, and other exciting stuff! Check out the job posting here: www.metacareers.com/jobs/3841154...
Research Scientist Intern, AI Research - Speech & Audio (PhD)
Meta's mission is to build the future of human connection and the technology that makes it possible.
www.metacareers.com
November 22, 2024 at 3:41 AM
Reposted by Desh Raj
Putting together a JHU Center for Language and Speech Processing starter pack!

Please reply or DM me if you're doing research at CLSP and would like to be added - I'm still trying to find out which of us are on here so far.

go.bsky.app/JtWKca2
CLSP
Join the conversation
go.bsky.app
November 19, 2024 at 3:37 PM
Hi, I'm Desh! I work on speech and LLMs at Meta NY (but you may also know me from my H1B rant going viral).
November 19, 2024 at 1:11 PM