PhD student at @uoft | NLP Enthusiast, Interested in Information Retrieval
Toronto, ON
Joined September 2021
- Tweets84
- Following183
- Followers120
- Likes530
Sajad Ebrahimi @ EMNLP 2026 retweeted
@sadjadeb is presenting a nice work from Reviewerly on peer review claim verification now in Hall H poster 103 at #EMNLP2026
Come say hi is you are working on topics related to peer review!
I am presenting "Peerify: benchmarking Peer Review Claim Verification" today at Hall H at #EMNLP2026
If you are working on problems related to peer review come visit us at poster 103!
Heading to EMNLP in Budapest and CIKM in Rome to present works on Peer Review Evaluation, Agentic Retrieval, RAG and Detecting AI-generated content.
Sessions and poster IDs below. If you’re curious about any of these, drop by and let’s chat!
#EMNLP2026 #CIKM2026
Sajad Ebrahimi @ EMNLP 2026 retweeted
Thanks @omarsar0 for sharing our paper before we even got the chance to! 🙏
Loved your framing so much we borrowed it for our thread
"same retriever, same agent loop, just a better opening move. ♛ "
Impressive paper showing how much the first retrieval step matters for deep research agents.
It helps to improve GPT-5.5 from 83.1% to 90.5% on BrowseComp-Plus with the same retriever and the same agent loop.
It seems that the gain comes from the opening context.
The authors propose Question's Gambit which runs once, before the agent starts searching.
It splits the question into clues, turns each clue into complementary searches, pools the results, and reranks them.
The agent then starts its loop with that ranked set already in context.
The same change lifts GPT-5.4-mini from 68.1% to 79.0% and DeepSeek-v4-pro from 71.4% to 76.9%, and roughly halves calibration error for GPT-5.5. It costs between 2.3 and 5.3 extra tool calls per question.
In an error analysis, only 3 of the 79 remaining GPT-5.5 errors come from the gold document never being retrieved. The other 76 happen later, when the agent previews, opens or uses the evidence.
Paper: arxiv.org/abs/2609.14412
Chat with Paper: academy.dair.ai/papers/quest…
Sajad Ebrahimi @ EMNLP 2026 retweeted
Impressive paper showing how much the first retrieval step matters for deep research agents.
It helps to improve GPT-5.5 from 83.1% to 90.5% on BrowseComp-Plus with the same retriever and the same agent loop.
It seems that the gain comes from the opening context.
The authors propose Question's Gambit which runs once, before the agent starts searching.
It splits the question into clues, turns each clue into complementary searches, pools the results, and reranks them.
The agent then starts its loop with that ranked set already in context.
The same change lifts GPT-5.4-mini from 68.1% to 79.0% and DeepSeek-v4-pro from 71.4% to 76.9%, and roughly halves calibration error for GPT-5.5. It costs between 2.3 and 5.3 extra tool calls per question.
In an error analysis, only 3 of the 79 remaining GPT-5.5 errors come from the gold document never being retrieved. The other 76 happen later, when the agent previews, opens or uses the evidence.
Paper: arxiv.org/abs/2609.14412
Chat with Paper: academy.dair.ai/papers/quest…
Sajad Ebrahimi @ EMNLP 2026 retweeted
I gave GPT-6 Astra the only benchmark I care about.
The helicopter mission in GTA Vice City.
Sajad Ebrahimi @ EMNLP 2026 retweeted
Just reviewed at AAAI ‘27, wonder how many of these papers were written entirely by AI.
5+1(Industry Day Talk) works accepted at #CIKM! 🎉 More details coming soon!
Sajad Ebrahimi @ EMNLP 2026 retweeted
Reviewer assignment: paper as a query on a collection of authors
Conflict of interest detector:
app.reviewer.ly/app/peerclea…
Evaluating reviews:
app.reviewer.ly/app/peerisco…
Human-AI collaboration detector
#SIGIR2026 @NegarEmpr
Sajad Ebrahimi @ EMNLP 2026 retweeted
DetectZoo: A Unified Toolkit for AI-Generated Content Detection Across Text, Audio, and Image Modalities
Sajad Ebrahimi, Nima Jamali, Bardia Shirsalimian, Kelly McConvey, Wentao Zhang, …
arxiv.org/abs/2606.04205 [𝚌𝚜.𝙼𝙼 𝚌𝚜.𝙰𝙸 𝚌𝚜.𝙲𝙻 𝚌𝚜.𝙲𝚅 𝚌𝚜.𝙻𝙶 𝚌𝚜.𝚂𝙳]
Excited to announce that our paper “From Noise to Order: Learning to Rank via Denoising Diffusion” has been accepted to #ICTIR2026! 🎉📚
📄Paper: arxiv.org/pdf/2602.11453
💻 Code: github.com/sadjadeb/Diffusio…
5/ 📈 Key benefits:
✅ Learns a more robust relevance estimation function by modeling the joint distribution over features + labels
✅ Matches inference cost of discriminative models
✅ Outperforms discriminative baselines on standard LTR datasets in our experiments.
A huge thanks to my amazing co-authors:
@UnderdogGeek , @NegarEmpr , @StevenYuan666 , @Haolun_Wu0203 , @FattaneZ , and @ebrahim_bagheri for the incredible collaboration on this research.
Check out our implementation on GitHub: github.com/sadjadeb/Diffusio…
Sajad Ebrahimi @ EMNLP 2026 retweeted
1/ 🎉PeeriScope from Reviewer.ly has been accepted @TheWebConf!
How good is your review?
Drop your review, or paste an OpenReview URL and it tells you how good the review is, across 13 dimensions.
Try it now 👇
🔗 app.reviewer.ly/app/peerisco…
📜arxiv.org/html/2604.24071v1
Sajad Ebrahimi @ EMNLP 2026 retweeted
1/ Excited to share that PeerPrism from Reviewer.ly has been accepted at #SIGIR2026
A benchmark for reality of today’s peer review, where reviewers draft the reviews, LLMs refine it, and “was this written by AI?” is no longer a yes/no question.
arxiv.org/abs/2604.14513
