Debdeep Sanyal

313 posts

Debdeep Sanyal

Debdeep Sanyal

@debdeeplikesai

im just trying to figure out how this "thinking" thing works. COLM 2025 || EMNLP 2025 🇨🇳

Katılım Mart 2022
296 Takip Edilen39 Takipçiler
Debdeep Sanyal
Debdeep Sanyal@debdeeplikesai·
@eigenron i see you mentioned about a real use case of interp. detection of alignment faking for starters on your plate? the fact that LLMs are mathematically bound to hallucinate? do you need a few more?
English
0
0
0
11
Debdeep Sanyal
Debdeep Sanyal@debdeeplikesai·
@eigenron it's a good way of letting others know you know shit about the field. first of all, "understanding LLMs" isn't a thing, it's like saying i understand the Alps. you can attempt at understanding certain parts of it, empirically or mathematically, and find pieces to a puzzle.
English
1
0
0
104
eigenron
eigenron@eigenron·
mech interp is as useful for understanding LLMs as a flight manual is to birds
AVB@neural_avb

@ylecun @nxthompson To be fair the original Anth post was just some cool mech interp expts. This particular tweet sensationalized it a bit much haha. PS: wonder what your thoughts are on the field of mech interp?

English
15
4
106
17.2K
Debdeep Sanyal
Debdeep Sanyal@debdeeplikesai·
@YuhangZhou2 looking forward to see you on thursday! really liked your paper when it was out, would love to have a chat
English
0
0
1
47
Debdeep Sanyal
Debdeep Sanyal@debdeeplikesai·
@davidbau sir, are you by any chance going to be at EMNLP this year? would love to meet once and have a talk about your insights
English
0
0
0
30
Debdeep Sanyal
Debdeep Sanyal@debdeeplikesai·
@MikaStars39 is the Qwen team visiting EMNLP 2025? would really love to have a chat with the team behind my go to LLM
English
0
0
0
39
Debdeep Sanyal
Debdeep Sanyal@debdeeplikesai·
@va_a14 unfortunately your sources are incorrect. friend got hired (undergrad isn't over yet) based on open source track record and research profile.
English
0
0
0
10
Vaibhav Agarwal
Vaibhav Agarwal@va_a14·
DSA is still the only source of truth if you want to earn 40-50LPA as a fresher in the tech industry.
English
78
54
1.7K
479.4K
Debdeep Sanyal
Debdeep Sanyal@debdeeplikesai·
@iclr_conf hoping that ICLR comes up with a system which is better from what NeurIPS did this time.
English
0
0
1
394
ICLR
ICLR@iclr_conf·
The Abstract deadline is soon: 11:59pm, Sep 19 (Anywhere on Earth). Don’t wait until the last minute 😉 Best of luck to everyone submitting!
English
3
10
130
40.6K
Debdeep Sanyal
Debdeep Sanyal@debdeeplikesai·
@prajdabre gemini is my go to. i love ai studio, i love 2.5 pro, i live a better life for it.
English
0
0
1
16
Raj Dabre
Raj Dabre@prajdabre·
If you're from India and I had to convince you to switch from other models to Gemini, what would I have to do?
English
123
0
207
25.2K
Debdeep Sanyal
Debdeep Sanyal@debdeeplikesai·
@Tanmoy_Chak @emnlpmeeting sir, wanted to ask if papers for which "Oral Presentation" is showing on OpenReview, are the papers confirmed to be Oral (given that the camera ready paper is well done)?
English
1
0
1
46
Debdeep Sanyal
Debdeep Sanyal@debdeeplikesai·
@xwang_lk @WenhuChen is it the reviewer's call to hide the rating? just experienced something similar where the reviewer said they're satisfied with the rebuttal and then hid their rating. first time experiencing, no idea how it works.
English
0
0
0
69
Wenhu Chen
Wenhu Chen@WenhuChen·
I don't quite understand the logic of hiding the review scores during the rebuttal/discussion period.
English
7
1
122
12.3K
Debdeep Sanyal
Debdeep Sanyal@debdeeplikesai·
@abeirami honestly, may the world be filled with more people like you who give others hope, especially young ones like me. thank you ahmad
English
0
0
7
2.3K
Ahmad Beirami
Ahmad Beirami@abeirami·
Only 2 out of 120 papers in my NeurIPS SAC batch have all reviewer scores ≥4! Don't despair with the low scores and focus on writing a clear and concise rebuttal! Good luck!
English
6
4
193
60.7K
Aakash
Aakash@shinyzenith72·
@debdeeplikesai @canadavisa_com We need more ai conferences to take place in Asia. A huge chunk of ai research comes from Chinese and Indian scholars. Accessible visa systems are crucial.
English
1
0
1
38
EMNLP 2026
EMNLP 2026@emnlpmeeting·
@b_azarkhalili @Osama_Rakan_ @danadaeun Actually, a gap of about one week is typical. For the February cycle, for example, the meta-review deadline was April 10th, and they were not released until April 16th. For the December cycle, they were due on February 7th and released on the 14th.
English
3
1
5
1.7K
Debdeep Sanyal
Debdeep Sanyal@debdeeplikesai·
@adyantalamadhya @ReviewAcl that's great! given how infuential meta-reviews are in the total accpetance pipeline, i'm sure ACs understand the weight of it. ofcourse i have read about a few ACs who are unlike you, but hey thanks for your work man. people like you putting hopes in us!
English
0
0
1
107