Azmine Wasi 💫 ICLR2026

Excited to share that two of my papers have been accepted to ICLR'26 (@iclr_conf), including one first author paper! One of the most selective conferences in AI/ML. This is my third consecutive year at ICLR (2024, 2025, 2026) as an undergrad.

@bremen79 @roydanroy I agree. It never does.

@roydanroy I edited the response to the reviewer and sent a private message to AC. My prediction: nothing will happen

Update: I wrote to the authors of FamousPaper'18 that replied that our result is great and they are happy to see this finally settled. How would you convey this information to AC and Reviewers? Also, this year you can reply only once to Reviewers and we already did...

All the reviewers who rated 5 acknowledged the rebuttal. Most of the 4s did too. But most of the 2s and 3s didn’t. Feels like the ones who give low scores regardless of merit are often the same ones who don’t bother engaging. #ICML #icml2026

always.

@sethlazar @NeurIPSConf That's better. Thank you so much for the clarification!

@AzmineWasi @NeurIPSConf (PPT co-chair here) Acceptance standards and criteria this year will be aligned to other tracks and other CS conferences. There is no explicit target but you can expect it to be around the same as the other tracks.

The Position Paper Track is back at NeurIPS 2026 for the second year, with an expanded scope, and better alignment with the main and Evaluation and Dataset tracks! Head to the Call for Paper at neurips.cc/Conferences/20… for all the important dates and information and read our accompanying blog post at blog.neurips.cc/2026/03/30/wha… to learn more about the changes we are making this year and how we adapted the process based on the feedback we got from the community! The submission deadline is the same as for the main and ED track: May 6, 2026 AoE. We are looking forward to read your papers and any feedback you may have!

@anshulkundaje @iScienceLuvr Agreed. At the same time, I am not sure about journals too. My experience with journals is equaly bad. Nothing is peer-reviewed anymore, actually.

@iScienceLuvr I get ur point wrt the OP. But NeurIPS is rigorously peer reviewed? In which universe? For core AI/ML areas that might be true to some extent. Applied work in NeurIPS is highly enriched for complete garbage. Reviewer pool really lacks domain expertise in applied areas.

I get the sentiment of this tweet, but it just reminds me how different the AI field really is, where conferences like NeurIPS are rigorously peer-reviewed and as prestigious as the top scientific journals.

First auto paper generation, now auto rebuttals. Is this accelerating research or simply overburdening our academic systems? If AI writes the papers and AI reviews them -- do we believe there will be real scientific progress or just a meaningless loop of content creation? I worry for our future conferences...

So, two reviewers misunderstood our position (#ICML '26 position paper track), got offended, and rated very low. One reviewer understood, asked very interesting questions, and rated it 5. Not sure what to do now 🙂

@langer_han very common in *ACL venues. tried of this.

I find it weird that many reviewers pick on limitations and use that as weakness in paper when authors point to limitations explicitly !!

Well, it seems, the reviewers DID NOT USE 3, 4 (weak reject and weak accept) SPARINGLY, at all! 😂😂 #icml2026

@NeurIPSConf It is kinda useless if we can't officially present papers there. NerurIPS'25 had EurIPS, but official presentation (for proceedings) was limited to US/Mexico, I think? So, please accept satellite event presentations as official for it to be much more effective!

Following the success of the EurIPS and NeurIPS-Mexico City pilots in 2025, we are thrilled to announce two official NeurIPS 2026 satellite events for this year! These will be held in Paris, France and Atlanta, USA, respectively, running alongside the main venue in Sydney, Australia. Both satellite events will feature keynotes, oral and poster presentations of accepted NeurIPS 2026 papers, as well as workshops. We are planning tutorials, affinity events, and other elements for the satellite sites and we'll share more information as planning advances. Wherever you choose to join us, the entire NeurIPS organizing committee is working hard to deliver an outstanding experience for the whole community! neurips.cc

@henryzhongsc @giffmana I agree 💯/💯! I think, irresponsible policy B reviewers should be punished too, in some way, if they have the watermarks present.

Yeah man, the last NeurIPS gave out a ton of complimentary registrations too. The current ICML also allows virtual registration for accepted authors, which basically costs them nothing. And the fact that it hosts the author self-ranking questionnaires more and more seriously also suggests that additional AC attention is something they envision to be at their disposal (as the discrepancy between reviewer rates and self-ranking rates is the main signal for this isotonic score thing). These are all great incentives. There are a lot of interesting incentives and penalties that can be connected to a credit system. I really couldn't care less whether they actually adopt my design or not; I just want the general mechanism to be there, as I believe it is, by design, superior to just saying nice words in reviewer guidance (which people just gloss over) and threatening desk rejection (which can only affect a very small scale). And yes, lying is just cherry-on-top. And honestly if you are really using LLMs responsibly you sure won't fall for such prompt injections (which basically only works if you ask an LLM to write the review). The fact that they got caught here just means they have no proper respect for other people's submissions. (On this note I kinda wonder if Policy B PDFs are watermarked, and if so, how many reviews there are flagged. @thegautamkamath )

I think this is good. Before replying with "but ..." do read their blogpost, they did take a lot of care here to avoid false positives. Sadly, these hundreds of people are only the most egregiously careless ones. Sad reckoning for the community.

@henryzhongsc @giffmana I agree. ICML also said to reward top 25% with free registration and top 50% with discount. Also, the one key fact is, the rejected ones LIED, EXPLICITLY. They said clearly they didn’t use LLM but they did. And, these lies justified the desk rejections, I think.

Yeah, I read it; I think what ICML did here is great. Reviewers who fall for such simple prompt injections are almost 100% offloading their duties to LLMs and deserve the DR. I am just saying your second note regarding broader poor review practices not being addressed — "these hundreds of people are only the most egregiously careless ones" (which is true, as only 1% of reviews got flagged) — is likely not something suitable for DR, since it is too nuclear. My take is that we need hard punishment for super low-FPR actions (missing a review, getting caught by an AI watermark, etc.), but the scale and impact of these are inherently limited. Yet, we need to accept a higher FPR and impose milder penalties (like taking away few points under a credit system) to achieve influence at meaningful scale. And yes, if a credit system becomes a new rule, its point-based consequneces should be in effect from day 1; otherwise, it is just another set of pretty words in the reviewer guidance (and an even more confusing one), which has proven to be ineffective. Don't think we disagree on anything but lmk.

@avt_im How are they checking this? I wrote my reviews manually, but I’m worried they might get flagged as AI-generated. Should I test them? I feel like after working/chatting with LLMs for a while, you start to adopt some of their patterns subconsciously. But any deliberate use is bad.

Sad times as an ICML AC: I've just learned that a mid-career faculty whose work I know and respect is getting their papers rejected for agreeing to the no-LLM reviewing policy, and then violating it.

This is crazy! If it continues, ARR is destined to fall. I wish I never submitted there. Such a waste of time and effort, in both submission and rebuttals.

Anyone noticed the recent trend of #ACL @ReviewAcl reviewers decreasing the overall scores (everything else is perfect) after the author response period, without any explanations?

One reviewer asked to cite four of his completely unrelated papers, in an @SpringerNature journal. Four papers have one author in common; I've googled the name and found his Google Scholar, FB, Insta, LinkedIn and everything (matched paper affiliation too).

@sidgairo18 that's the hope.

@AzmineWasi Understandable. But being good and holding your values + integrity always helps in the long run. Plus you keep your conscience clear.

"The author's rebuttal addressed my concerns and I maintained my original score." Sigh! This is the most BS justification for final review scores. What then is the damn reason to maintain your scores (especially when the score is borderline) ? To me (imho) it's either the rebuttal answers your concerns and you update to an accept or it doesn't and then you lean towards a reject. And this was despite the fact that the reviewer instructions clearly stated not to have any "Borderline Accept / Reject" scores.

@langer_han @sidgairo18 One AC rejected our LLM-based Legal Agent paper (with rigorous expert evaluation) in EACL 2026, saying "out of scope". When judges are biased, the system is destined to collapse.

@sidgairo18 @AzmineWasi I have been ranting about this too. For ARR I kid you not I have weakness as this even though the other reviewers found in appropriate and clear (in NLP). I really hope borderline scores and reviewers need to be held accountable