Joined February 2023
- Tweets 47
- Following 117
- Followers 93
- Likes 31
8 Photos and videos
Pinned Tweet
Apr 15
New blogpost out π
"Detecting LLM Misbehaviors from the Inside Out with Deep Learning on Structured Data" (ffabffrasca.substack.com/p/dβ¦) [1/8]
1
6
10
1,935
1/
How much can you compress an LLMβs KV cache?
tl;dr it depends on how you train your model.
Many strong context compaction methods, such as Cartridges and attention matching, operate post-hoc: given a fixed model and a context, they try to compress the resulting KV cache.
@yoav_gelberg and I ask the complementary question:
can we train the model to produce KV representations that are easier to compress?
In other words: keep the compression method fixed, and change the representations it sees.
6
18
99
29,198
Apr 15
New blogpost out π
"Detecting LLM Misbehaviors from the Inside Out with Deep Learning on Structured Data" (ffabffrasca.substack.com/p/dβ¦) [1/8]
1
6
10
1,935
Apr 15
- "Neural Message-Passing on Attention Graphs for Hallucination Detection", ICLR 2026 (arxiv.org/pdf/2509.24770) [7/8]
1
1
79
Apr 15
- "Beyond Next Token Probabilities: Learnable, Fast Detection of Hallucinations and Data Contamination on LLM Output Distributions", AAAI 2026 (arxiv.org/pdf/2503.14043) [8/8]
51
Feb 19
Check out our new ICLR 2026 paper - we explore hallucination detection through graph learning. Take a look!
Feb 18
π§΅"Neural Message Passing on Attention Graphs for Hallucination Detection" at #ICLR2026 !
πΈοΈWe apply GNNs on the structured data LLMs produce as they generate text (e.g. attentions) to predict their errors.
π arxiv.org/abs/2509.24770
π€ @GuyBarSh (co-1st) @YftahZ @HaggaiMaron
5
17
1,400
Feb 2
π [4/4] On the Expressive Power of GNN Derivatives
We study how using gradients of GNNs can increase their expressive power, providing a principled way to go beyond standard message passing.
arxiv.org/pdf/2510.02565
1
1
4
307
Feb 2
These works were joint efforts with a group of amazing collaborators: @ffabffrasca @HaggaiMaron @YftahZ @ytn_ym @yaniv_galron @yoav_gelberg @itayevron @mayabechlerspei @ido_guy Ami Tavory Moshe Eliasof Ran Elbaz
2
5
347
Guy Bar-Shalom retweeted
30 Nov 2025
π Beyond Token Probes: Hallucination Detection via Activation Tensors with ACT-ViT
w/ @GuyBarSh , @ffabffrasca, Yaniv Galron, @YftahZ
arxiv.org/abs/2510.00296
2
9
670
Guy Bar-Shalom retweeted
5 Dec 2025
@GuyBarSh and I will be presenting the poster today, stop by π€
π Fri, Dec 5 β’ 4:30β7:30 PM PST β’ Exhibit Hall C,D,E # 4000
4 Oct 2025
[1/7] New paper: "Beyond Token Probes: Hallucination Detection via Activation Tensors with ACT-ViT" #NeurIPS2025 [arxiv.org/pdf/2510.00296]
Joint work with: @ffabffrasca (co-first), @yaniv_galron, @YftahZ , @HaggaiMaron
3
7
899
Guy Bar-Shalom retweeted
27 Nov 2025
π€ Can discrete diffusion models actually outperform standard classifiers?
We show that it can!
π arxiv.org/pdf/2511.20263
π» github.com/omerb01/didicm
π omerb01.github.io/didicm-web
1
4
17
7,343
4 Oct 2025
[1/7] New paper: "Beyond Token Probes: Hallucination Detection via Activation Tensors with ACT-ViT" #NeurIPS2025 [arxiv.org/pdf/2510.00296]
Joint work with: @ffabffrasca (co-first), @yaniv_galron, @YftahZ , @HaggaiMaron
1
3
16
4,044
4 Oct 2025
[6/7] Results (over 15 LLM/dataset combinations):
β’ Consistently outperforms classic probes
β’ Zero-shot generalization to new datasets
β’ Fast adaptation to unseen LLMs by tuning only their new corresponding adapter
1
1
84