Harman Singh (@Harman26Singh)

Pinned

H

Harman Singh@Harman26Singh · Jun 25

🚨 New @GoogleDeepMind paper 𝐑𝐨𝐛𝐮𝐬𝐭 𝐑𝐞𝐰𝐚𝐫𝐝 𝐌𝐨𝐝𝐞𝐥𝐢𝐧𝐠 𝐯𝐢𝐚 𝐂𝐚𝐮𝐬𝐚𝐥 𝐑𝐮𝐛𝐫𝐢𝐜𝐬 📑 👉 arxiv.org/abs/2506.16507 We tackle reward hacking—when RMs latch onto spurious cues (e.g. length, style) instead of true quality. #RLAIF #CausalInference 🧵⬇️

AAK@_akhaliq · Jun 24

Robust Reward Modeling via Causal Rubrics

4

31

123

80

25.0K

Harman Singh Retweeted

P

Partha Talukdar (✈️ ACL 25)@partha_p_t · 11 h

At #GoogleIOConnect, we demoed a Gemini-powered research prototype offering mixed-language step-by-step guidance. youtube.com/live/is0n2n-x4… (starting 6:40 for 2mins) We envision graduating such capabilities in education & other domains to Gemini Live, impacting users across…

2

3

42

8

2.0K

H

Harman Singh@Harman26Singh · Jul 24

Opportunity to work with some amazing researchers and touch the lives of billions of people across the world.

PPartha Talukdar (✈️ ACL 25)@partha_p_t · Jul 23

@GoogleDeepMind India 🇮🇳 & Japan 🇯🇵 are looking for strong candidates in multilinguality, multicultural, & multimodality areas. RS Bangalore: job-boards.greenhouse.io/deepmind/jobs/… RS Tokyo: job-boards.greenhouse.io/deepmind/jobs/… RE Tokyo: job-boards.greenhouse.io/deepmind/jobs/…

0

2

43

4

3.0K

Harman Singh Retweeted

O

Owain Evans@OwainEvans_UK · Jul 22

Our setup: 1. A “teacher” model is finetuned to have a trait (e.g. liking owls) and generates an unrelated dataset (e.g. numbers, code, math) 2. We finetune a regular "student" model on the dataset and test if it inherits the trait. This works for various animals.

5

47

1.0K

122

79.0K

H

Harman Singh@Harman26Singh · Jul 23

Great opportunity to work with @partha_p_t @heiga_zen and many others in GDM on multilinguality, multicultural, & multimodality for Gemini Application links below ⬇️⬇️

PPartha Talukdar (✈️ ACL 25)@partha_p_t · Jul 23

@GoogleDeepMind India 🇮🇳 & Japan 🇯🇵 are looking for strong candidates in multilinguality, multicultural, & multimodality areas. RS Bangalore: job-boards.greenhouse.io/deepmind/jobs/… RS Tokyo: job-boards.greenhouse.io/deepmind/jobs/… RE Tokyo: job-boards.greenhouse.io/deepmind/jobs/…

0

1

18

3

1.0K

Harman Singh Retweeted

S

Sumanth@sumanthd17 · Jul 23

Got a chance to showcase what we’ve been building at @SarvamAI to a larger audience at @google I/O Connect. Thanks for having us @osanseviero, was great sharing what we’re up to! (so far) 🚀

4

10

227

11

9.0K

H

Harman Singh@Harman26Singh · Jul 23

Friends doing incredible things. Congrats @sumanthd17 😄!!

SSumanth@sumanthd17 · Jul 23

Got a chance to showcase what we’ve been building at @SarvamAI to a larger audience at @google I/O Connect. Thanks for having us @osanseviero, was great sharing what we’re up to! (so far) 🚀

0

8

0

448

H

Harman Singh@Harman26Singh · Jul 22

Very excited to announce that I’ll be co-organizing a @NeurIPSConf workshop on LLM evals! Identifying shortcomings in model capabilities in a robust, scientific way is a critical part of model development. Looking forward to discussing ideas and hearing from some eval experts!

LLLM Evals Workshop @NeurIPS@LLM_eval · Jul 22

We are happy to announce our @NeurIPSConf workshop on LLM evaluations! Mastering LLM evaluation is no longer optional -- it's fundamental to building reliable models. We'll tackle the field's most pressing evaluation challenges. For details: sites.google.com/corp/view/llm-…. 1/3

2

11

65

4

7.0K

Harman Singh Retweeted

S

Stella Li@StellaLisy · Jul 22

WHY do you prefer something over another? Reward models treat preference as a black-box😶‍🌫️but human brains🧠decompose decisions into hidden attributes We built the first system to mirror how people really make decisions in our #COLM2025 paper🎨PrefPalette✨ Why it matters👉🏻🧵

6

69

361

253

39.0K

H

Harman Singh@Harman26Singh · Jul 22

Novel cognitive science grounded approach to preference modeling: synthetic counterfactual training + attention-based attribute integration. Empirical validation across 45 communities with human evaluation confirming interpretability claims! 🌟

SStella Li@StellaLisy · Jul 22

WHY do you prefer something over another? Reward models treat preference as a black-box😶‍🌫️but human brains🧠decompose decisions into hidden attributes We built the first system to mirror how people really make decisions in our #COLM2025 paper🎨PrefPalette✨ Why it matters👉🏻🧵

0

3

7

1

811

Harman Singh Retweeted

P

Partha Talukdar (✈️ ACL 25)@partha_p_t · Jul 23

Fantastic opportunities to contribute to Gemini through foundational work and touch billions! Please DM me and @heiga_zen for any clarification I shall also be at @aclmeeting 🇦🇹, happy to chat in person!

0

6

17

2

5.0K

Harman Singh Retweeted

P

Partha Talukdar (✈️ ACL 25)@partha_p_t · Jul 23

@GoogleDeepMind India 🇮🇳 & Japan 🇯🇵 are looking for strong candidates in multilinguality, multicultural, & multimodality areas. RS Bangalore: job-boards.greenhouse.io/deepmind/jobs/… RS Tokyo: job-boards.greenhouse.io/deepmind/jobs/… RE Tokyo: job-boards.greenhouse.io/deepmind/jobs/…

2

23

142

74

61.0K

H

Harman Singh@Harman26Singh · Jul 23

Checkout the new general purpose audio encoder trained by @ShikharSSU !

SShikhar@ShikharSSU · Jul 22

Meows, music, murmurs and more! We train a general purpose audio encoder and open source the code, checkpoints and evaluation toolkit.

0

1

5

1

509

H

Harman Singh@Harman26Singh · Jul 22

Meows, music, murmurs and more! We train a general purpose audio encoder and open source the code, checkpoints and evaluation toolkit.

aarXiv Sound@ArxivSound · Jul 21

Shikhar Bharadwaj, Samuele Cornell, Kwanghee Choi, Satoru Fukayama, Hye-jin Shim, Soham Deshmukh, Shinji Watanabe, "OpenBEATs: A Fully Open-Source General-Purpose Audio Encoder," arxiv.org/abs/2507.14129

0

15

32

4

4.0K

Harman Singh Retweeted

J

Jeff Dean@JeffDean · Jul 21

Congrats to the whole Gemini team, and especially to those focused on our advanced reasoning and mathematics capabilities! 🎊 More details in the blog post: deepmind.google/discover/blog/…

2

3

176

9

24.0K

Harman Singh Retweeted

Y

Yuchen Jin@Yuchenj_UW · Jul 21

I love this gang.

2

1

197

5

12.0K

Harman Singh Retweeted

k

koray kavukcuoglu@koraykv · Jul 21

Advanced version of Gemini Deep Think (announced at #GoogleIO) using parallel inference time computation achieved gold-medal performance at IMO, solving 5/6 problems with rigorous proofs as verified by official IMO judges! Congrats to all involved! deepmind.google/discover/blog/…

30

155

756

71

97.0K

H

Harman Singh@Harman26Singh · Jul 20

Thrilled to have contributed to Gemini 2.5 Pro with @partha_p_t and many other folks from GDM 🚀🚀🚀 keep a lookout for hiring apps for GDM India!

PPartha Talukdar (✈️ ACL 25)@partha_p_t · Jul 19

It's extremely gratifying to see so many contributors to Gemini 2.5 from @GoogleDeepMind India! Sip filter coffee (w/ plant-based milk, of course) as you pave the path to AGI, can't think of a better deal 🙂 (Btw, we are growing, app link coming soon!) arxiv.org/abs/2507.06261

0

3

29

2

2.0K

H

Harman Singh@Harman26Singh · Jul 21

Happy to see wonderful contributions from our team, making Gemini 2.5 models more powerful, more efficient, and understand many more languages and cultures!

PPartha Talukdar (✈️ ACL 25)@partha_p_t · Jul 19

It's extremely gratifying to see so many contributors to Gemini 2.5 from @GoogleDeepMind India! Sip filter coffee (w/ plant-based milk, of course) as you pave the path to AGI, can't think of a better deal 🙂 (Btw, we are growing, app link coming soon!) arxiv.org/abs/2507.06261

1

2

41

1

3.0K

Harman Singh Retweeted

E

Eddy Quan@waronweakness · Jul 20

Set a deadline

47

923

6.0K

4.0K

316.0K