PhD CS @TUDelft | Advised by @chiragraman | Generative Multimodal ML | Free Energy | Cognition | Physical Intelligence | https://teapaw.org/t.co/3Ds5iS4gQm
Delft, The Netherlands
Joined August 2009
- Tweets1.8K
- Following1.7K
- Followers161
- Likes76.7K
Ojas Shirekar retweeted
ICLR 2027 reviews are out. I pulled every ICLR review since 2025, about 250k of them, and it's not looking great.
AI written reviews are now the majority (* based on a sample):
2025: 18%
2026: 41%
2027: 65%
Scores are collapsing into the middle, and as others have pointed out - most consolidate around "Weak reject" (score 2). Weak rejects went from 36% of all scores to 57%. Clear accepts and rejects fell from 41% to 21%.
An indication that this is an AI phenomena - AI-flagged reviews picked a clear side less than half as often as humans.
Then you have the really awkward slip-ups.
94 reviews this year call the paper "the provided manuscript." That's chatbot for "the PDF you uploaded." In 2025 there were zero.
28 complain the paper is full of garbled numbers. Those are probably the line numbers embedded in the submission's margins, which is what they look like after a chatbot turns the PDF into text.
23 reviews have a Weaknesses section written for a different paper, word for word. For example, one paper on RL training got criticized for its chemistry. It has no chemistry. In half of these cases, the two reviews were posted less than 9 minutes apart.
I'll leave everyone to draw their own conclusions, but clearly we're not adapting quickly enough.
Ojas Shirekar retweeted
I am an AC for ICLR 2026. One of the papers in my batch was just withdrawn. The authors wrote a brief response, explaining why the reviewers failed at their job. I agree with most of their comments. The authors gave up. They are fed up. Just like many of us. I understand. We pretend the emperor has clothes, but he is naked.
Here is the final part of their withdrawal notice. I took the liberty to make it public, to highlight that what we are doing with AI conference reviews these last few years is, basically, madness.
---
Comment: We thank the reviewers for their time.
However, upon reading the reviews for our paper, it became immediately apparent that the four "reject" ratings are not based on good-faith academic disagreement, but on a critical failure to read the submitted paper.
The reviews are rife with demonstrably false claims that are directly contradicted by the text. The core justifications for rejection rely on asserting that key components are "missing" when they are explicitly detailed in the manuscript. Some specific examples are (and many are even fake claims).
Claim: Harder tasks like GSM8K are missing.
Fact: GSM8K results are in many tables, like Table 2 (Section 4.2) and Appendix G.
Claim: The method does not use per-layer ranks.
Fact: This is the entire point of our method. The reviewer clearly mistook our method for the baselines. (Section 2, Table 1).
Claim: The GP kernel is not specified.
Fact: It is specified in Appendix E (Table 6).
Claim: There is no ablation of the method's three stages.
Fact: Section 4.4 ("Ablation Study") and Appendix J are dedicated to this.
Reviewers have a fundamental responsibility to read and evaluate the work they are assigned. The nature of these errors is so fundamental, so systemic in overlooking explicit content, that it goes far beyond what "limited time" or "oversight" can explain. This work has gone through several rounds of revision over the last year. In earlier submissions, the paper usually received borderline or weak-accept scores.
Numerous signs strongly suggest that some reviewers are relying entirely on AI tools to automatically generate peer reviews, rather than fulfilling their fundamental responsibility of personally reading and evaluating manuscripts.
We strongly protest this.
This is a gross disrespect to the authors. It is a flagrant desecration of the reviewer's sacred duty. It fundamentally undermines the integrity of the entire peer-review process.
Given that the reviews are not based on the actual content of our paper, we have decided to withdraw the submission.
We leave this comment so that future readers of the OpenReview page are aware that the items described as "missing" are already present in the submitted manuscript. These negative reviews for this submission are factually unsound and do not reflect the content of the paper. We cannot and will not accept an assessment that is not based on the work we actually submitted.
Ojas Shirekar retweeted
We suddenly became twice as interested in the "appendix"... 🙃 #ICLR2027
Couldn’t have said it any better. I’m unsure if reviewers deliberately ignore the distinction. I now have two reviewers who think me citing the arxiv version of some papers is a critical weakness 🤷🏻♂️
One of the more interesting benchmarks I have seen lately!
Introducing ShadowBench: can frontier models recreate shadow puppets?
The model gets an RL environment with a curtain and 3D hands, and must pose them joint by joint until it matches a diagram from a 150 year old (!) puppet book.
Crafted at @perpetual_si
Happy to announce our new pre-print on Multiparty Human Motion generation!
Paper: arxiv.org/abs/2609.37708
Most generative models of human motion treat people independently or leave this structure implicit. Our new preprint makes it explicit.
BRAID (Bilevel Representations for Agent Interaction Dynamics) is a hierarchical sequential latent-variable model of multi-person motion:
→ A group latent captures the shared dynamics of an interaction; person latents capture each individual's behaviour, conditioned on that evolving context.
→ Real social scenes are rarely fully visible, because people occlude themselves and each other. BRAID is a hierarchical sequential Neural Process, so its latents are inferred from whatever subset of joints and people is observed.
→ One model with no task-specific heads handles social forecasting, inpainting of occluded joints or entirely missing people, and partner response generation. It is trained on SMPL-X bodies across conversation, dance and boxing data.
→ It produces varied plausible completions, not a single averaged future. A socially valid continuation need not match the one recorded future, so alongside standard metrics we evaluate synchrony and cross-person structure.
→ The latents are inspectable: for instance group latents separate Latin from Western ballroom styles, and person latents keep the same individual close across different groups without explicit identity features.
Why it matters:
- For motion capture, this gives interaction-aware priors, since the people around an occluded person constrain what they are plausibly doing.
- For embodied agents, it gives a compact social state: how the interaction is unfolding and how each participant tends to behave.
- Can drive socially coherent digital humans, act as a social simulator for robots planning around people,
- Or feed world models/LLMs/VLMs the social state vectors to help them reason about people.
Work done with @chiragraman, Yash Surange and Augustinas Jucas (@tudelft @EEMCS_TUD @TUDelft_AI)
Ojas Shirekar retweeted
I took an Advanced Topics in Computer Vision for my masters in CS focused in AI. CSC 577. Top-tier American university, 2017.
I was the only non-Chinese student in the class. The professor was Chinese.
I arrived late for a class by ~ten minutes, walked in, and it was being taught in Mandarin.
This debate is insane
Insane. Can soon have custom neural nets with weights edited by other models.
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
Ojas Shirekar retweeted
Be like 🇨🇭
Don’t be like 🇪🇺
Arnaud nails it. ✅👇
Other European countries should consider studying the foreign policy of their Swiss neighbor: they've achieved the pretty extraordinary feat of securing duty-free access to China for over 99% of their exports (the final number is actually 99.8%: caixinglobal.com/2026-08-25/…).
Funny what happens when you're a genuine democracy that works for its citizens' interests and does true diplomacy: much less theatrics and a lot more to show for it.
Ojas Shirekar retweeted
I just got back from India. It was the most incredible trip, filled with renewed relationships, new friendships and new knowledge.
There is no doubt traveling is hard, but what we gain from expanding our perspectives outweighs any biological wreckage we might experience.
Ojas Shirekar retweeted
Can someone please train a frontier language models that is actually good at LANGUAGE! Specifically writing. WTF is wrong with these shitty ass frontier models with absolutely horrible writing. They seem to be getting worse with every version.
Ojas Shirekar retweeted
What a bizarre future. I wanted to play Halo: Combat Evolved for the first time with @robertskmiles, and we wanted local multiplayer across two screens, and a model could just immediately add that feature to the decompilation.
Ojas Shirekar retweeted
How can AI systems explore combinations of ideas that they have rarely, or never, encountered?
In PRISM, we use analogical reasoning to support creative exploration in the long tail of the data distribution: drawing on familiar relationships to make sense of unfamiliar combinations. Specifically, we introduce an approach grounded in category theory to make conceptual relationships explicit, measure their correspondence across domains, and use them to guide generation.
We use visual metaphor generation as a concrete test. What would it mean to depict "Thought is a vulture" or "My mind is a desert and I am an explorer"? The images need to convey the relationships that make these analogies meaningful.
The gallery compares zero-shot generation with PRISM's refined outputs. Both PRISM variants use GPT-Image 1.5 to generate images, with Claude Sonnet 4.6 or GPT-6 Astra extracting relational structure and guiding prompt revision.
Work led by @mirellazeisler, with @ossjs, @MirceaTudor4, and me at the Tapri Lab.
📄: arxiv.org/abs/2610.01383
#AnalogicalReasoning #Creativity
🧵1/5
Ojas Shirekar retweeted
Elon, you've officially run out of arguments. Now every Indian who disagrees with you is "Ambani's shill."
Nobody said Starlink is too strong. Nobody's scared of your competition. India's problem is you. A Pentagon contractor building Trump's Golden Dome wants to run internet over Indian soil, and he throws a tantrum every time someone asks who holds the off switch.
Ambani's de facto monopoly?
You SIGNED A DEAL with that monopoly in March 2025 to sell Starlink in India. You didn't have a problem with Ambani when he was your reseller.
You only discovered his monopoly when Jio announced its own satellites.
And let's talk about hypocrisy.
Tesla's biggest factory is in Shanghai. Starlink is banned in China. Have you ever called Xi Jinping a monopolist?
Called Chinese critics shills? Written a sarcastic letter to Prime Minister Xi?
No.
Because Beijing would shut your factory by Monday. India can't, so India gets the insults.
now the math your own fans keep posting starlink could add 100–200 million Indian homes. Starlink is approved for 600 Gbps over India. Split across 100 million homes, that's 6 kilobits per second each. Your revolution is slower than a 1998 dial-up modem. 🤡
If it's too expensive, no one will buy it.
Your own India website showed ₹8,600 a month. A Jio plan costs ₹349. Indians already ran the numbers.
Five days of meltdown. Crime against India. Prime Minister Ambani. collages. And now shills.
The richest man on earth, crying on his own app because a sovereign country wants a security check.
Sit down, elon. Finish the paperwork. India isn't a tesla shareholder you can bully on X.
The ridiculous arguments by Ambani’s army of Internet shills make no sense
On the one hand, they say Starlink couldn’t possibly compete with the amazing low prices offered by Ambani’s de facto monopoly in India and then, in the same breath, they essentially claim that Starlink would be too strong of a competitor! This is absurd hypocrisy and hyperbole.
The reality is that competition would greatly benefit the average citizen in India by giving them more choices, as it always does. Ambani’s profits will be lower, but he is not going out of business. And, obviously, if Starlink is too expensive in India, no one will buy it, so SpaceX will need to make pricing affordable.
Ojas Shirekar retweeted
Imagine referring to someone solving problems in a field as the field being annihilated...
Fields medallist Hugo Duminil-Copin was deeply shaken by OpenAI’s release of more than 700 AI-generated mathematical manuscripts, including work on 16 questions directly relevant to his research.
“On Tuesday night, I cried, and it woke up my daughter”
“They have annihilated my field of research.”
lemonde.fr/sciences/article/…
Ojas Shirekar retweeted
Europe is dead.
My son hurt his foot. Couldn't sleep from the pain. Called for help: give painkillers, call tomorrow.
Now 4 hours at emergency in the Netherlands, still waiting for a cast.
Taxed 50% on everything. This is what you get when you need care.
Time to leave.
Ojas Shirekar retweeted
Dear Elon @elonmusk
You hv certainly stirred up plenty of reaction with ur posts. But this needs a response from me, as did ur “Electronic Voting machines can be hacked” post from some time ago.
As someone who knows a little bit abt Telecom as someone who played a decent-ish role in building Indias wireless sector as an Entrepreneur and then as a policy maker , let me share my thoughts.
It is true that Indias Telecom was a sector beset with opaqueness and corruption for several decades - where cronies ruled under the benevolent love and the politics of ur latest fan (name not withheld - @RahulGandhi )
But under PM @narendramodi , 3 principles govern the tech (including Telecom/Internet) sector.
Open, Safe & Trusted and Accountable.
#OSTA
✅ Open - Free from any monopoly-duopoly. Free and fair competition.
✅ #SafeAndTrusted - Infra, Network , Platforms must be safe and trusted.
✅ Accountable - Every participant should be legally accountable for safety and trust.
I was a MP first and then part of PM @narendramodi jis team when this framework was developed. I have more than a little experience of fighting years of political and bureaucratic opaqueness and corruption, which shaped these principles.
so todays India is not that old crony driven India.
I am not in govt anymore and certainly dont speak for it. But it seems to me that Govt has made it clear: all three satcom licensees are at broadly the same stage of ensuring security clearance by verifying its safety and trust first, then spectrum.
Satellite IP services have many security implications. More so in a country like India which has faced years of Terror and insurgency supported by inmical state actors in our neighbourhood. From 26/11 terror attacks to recent ones in Pahalgam, satcom has been preferred comm for Terrorists. Satellite VOIP and encryption creates more challenges security wise. Not insurmountable but new capabilities and capacity hv to be built to ensure Safety and Trust.
I hv used @Starlink and I think its awesome technology - and represents a new era in mobility innovation. No doubt.
But a country has to plan for itself and its interests before every new choice. Thats what India is doing. Patience is unavoidable and in this case is a necessity.