Azmine Wasi ✈️ NeurIPS'26

AI/ML Research (GenAI:LLMs/VLMs, Agents and Reasoning, MedAI, HAI-Safety) 📝 Explorer ⚙️ Seeking Research Intern/Fellowship Opportunities ⏳Views are mine

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

+3 in ICML!
1 Main
2 Position Papers (1 Spotlight)!

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

@murat_kocaoglu_ I'm afraid we have AI ACs too....

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

@shyoyhs @mamoruk surely! 🥺
that's so nice of them! 😁😆

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

In AACL-IJCNLP 2026, "LLM-Ideoplasticity: Measuring Ideological Plasticity in the Political Behavior of LLMs as a Context-Conditioned Distribution" won the best paper award.

Funny part? The EXACT same paper (with avg 4.17) was REJECTED from EMNLP! Same ARR cycle, no improvement!

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

@quarbby Yes, it happened to me multiple times too!

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

@zehavoc Exactly.

If one AC/SAC disliking your paper can create this much gap (reject vs award), obviously, its time to rethink about power concentration and accountability gap in the higher layers of program committee.

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

@prajdabre That's the part I was talking about.
Surely, one AC/SAC dismissed it in EMNLP; another one embraced it in AACL.
But, even if it happened, the gap between these two decisions are too big to ignore the accountability gap here.

[Mathieu @miniapeur]

LLMs are authors. LLMs are reviewers.

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

It just shows how MESSED up the current review+adjudication system is.

Is the best paper of AACL-IJCNLP is worse than the worst EMNLP paper? Nope!

It shows that you need to have good luck, not a good paper. Award-worthy stuffs may not even be in top 5K, if you're unlucky.

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

Anyone Pangrammed all the reviews like 2026?

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

Famous words from Linkin Park about NeurIPS-ICLR-ICML:

"We're building it up, to break it back down:
We're building it up, to burn it down:
We can't wait, to burn it to the ground...." 😆

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

@_sungmin_cha Same :)

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

AI-proofing your work works.

[Arth Singh 🦉 @iarthsingh]

Guys this ICLR paper got a 4 ? WTF ??? Generational AI slop it was

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

Exactly!

[Jiaxin Pei @jiaxin_pei]

My biggest struggle with reviews in 2026 is that (AI) reviewers don’t seem to know the difference between limitations and critical issues. An LLM can easily find 100+ limitations in any paper (even the best ones in history), but that doesn’t mean the paper shouldn’t be accepted.

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

Exactly.

Reviews are bad (alreadt); I bet 70% of the reviewers won't even care to respond and 90% won't increase scores either (despite good rebuttal); as the scale is "imbalanced".

Comparing to 1-10:
1-3: 1
4-5: 2
6: 3
7-10: 4

[Yanlin Fu @ylfu02]

4分值不就是纯纯拼谁运气好拿到好心审稿人吗😅

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

ICLR 2027 is a mess.

The reviews, the questions, the ratings, even the confidence scores are f****d....

Everything is upto the ACs and SACs. I hope and pray they'll work diligently and do justice.

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

@imbaban00 Not sure.

I have seen 3rd/4th/5th year PhD students before, in 2025, so, PhD completion is not required maybe.
Enough publication and reviewing experience seems fine for it.

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

@paahrsa @iclr_conf I got something similar.
They invented a problem doesn’t exist and then provided a solution that doesn’t work. 🤷‍♂️

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

I don't think any human will rate 4 too (from past experience). After initial review, I doubt if there will be 100+ 4s.

Cause, 4 is like previous 7-10 rating; and the papers deserving 7 is not likely to get rated 4, as it inflates the score to a 8-10 level; which is very rare.

[Jaron Maene @jjcmoon]

I let gpt-6 astra review 250 random ICLR submissions. It didn't recommend clear accept for a single paper.

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

@ChaiKamNamak @aks1812 That's very true.

I think 2-3 papers per person capping can be a good idea (maybe except AISTATS-style resubmissions).

Reasonable gatekeeping is very fine to me. But mass desk-rejection for weak, un-scientific reasons - I'm not comfortable here.

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

@hbaniecki @iclr_conf And, don’t wanna drag with further. It was nice discussing with you!

If claude-assisted papers are not good enough (considering all AI-slop tweets), I don't accept that they are capable of such deep, informative and useful reviews either.

Best wishes for you!

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

It seems some @iclr_conf reviewers are using Pangram or such detectors and possibly, introducing some sort of bias, based on detector output.
But, as far as I know, reviewers are not allowed to copy-paste any text to such LLM/AI tools, except local inference.

[Azmine Wasi ✈️ NeurIPS'26 @AzmineWasi]

@hbaniecki @iclr_conf Sorry, I disagree with you here.

I mean, if you believe you are not capable enough to review the paper properly without such direct assistance, just message AC to reassign something that relates with you more. Simple.