Omri Fahn (@OmriFahn) header
Omri Fahn (@OmriFahn) profile picture

Omri Fahn

@OmriFahn
Joined February 2022
  • Tweets24
  • Following55
  • Followers16
2Media
Photo by Omri Fahn (@OmriFahn)Photo by Omri Fahn (@OmriFahn)
Omri Fahn retweeted
Mor Geva (@megamor2) profile picture
Mor Geva
@megamor2May 1
3 papers accepted at #ICML2026 🇰🇷🎉 🔸 From Directions to Regions: Decomposing Activations in Language Models via Local Geometry arxiv.org/abs/2602.02464 🔸 Interpretability can be Actionable actionable-interpretability-… 🔸Hallucinations Undermine Trust; Metacognition is a Way Forward (out soon) @_galyo @ymatias Congrats to all collaborators!
From Directions to Regions: Decomposing Activations in Language...

From Directions to Regions: Decomposing Activations in Language...

Activation decomposition methods in language models are tightly coupled to geometric assumptions on how concepts are realized in activation space. Existing approaches search for individual global...

arxiv.org
8
11
2
176
10,128
MFA is a beautiful visualization, but not only that. It’s a practical tool: competitive localization and better steering, while revealing that concepts live in regions, not just single directions. Grateful I got to contribute to this with an awesome team!
Or Shafran (@OrShafran) profile pictureOr Shafran@OrShafranFeb 5
It's time to look past dictionary learning for decomposing LM activations. What happens when we instead leverage local geometry? We find a natural region-based decomposition that yields better steering and localization 🧵 1/
1
1
2
81
Honored to see Formula Completion highlighted by @satyanadella 🎉 Proud to be part of the incredible team bringing this to life at Microsoft Excel!
Satya Nadella (@satyanadella) profile pictureSatya Nadella
@satyanadella14 Oct 2025
This is super useful! With new Formula Completions in Excel, just type "=" and Copilot proactively suggests a formula, based on the context of your sheet. Here's a great example.
1
1
122
This is super useful! With new Formula Completions in Excel, just type "=" and Copilot proactively suggests a formula, based on the context of your sheet. Here's a great example.
122
315
76
2,189
443,221
🤔 Can an LLM sense it's about to hallucinate—before it speaks? 🤔 And do those gut feelings live inside its geometry? Thrilled to share our new paper: “Pre-trained LLMs Learn Multiple Types of Uncertainty” (Cohen • Fahn • de Melo, 2025) 📄 arxiv.org/abs/2505.21218
Pretrained LLMs Learn Multiple Types of Uncertainty

Pretrained LLMs Learn Multiple Types of Uncertainty

Large Language Models are known to capture real-world knowledge, allowing them to excel in many downstream tasks. Despite recent advances, these models are still prone to what are commonly known...

arxiv.org
1
34
🤔Can an LLM "unconscious" feel a lie bubbling up before it speaks? 🤔And what mysteries hide in its linear space of uncertainty? Thrilled to share our new paper: “Pre-trained LLMs Learn Multiple Types of Uncertainty” (@roicohen9•@OmriFahn•@GerardDeMelo) arxiv.org/abs/2505.21218
Pretrained LLMs Learn Multiple Types of Uncertainty

Pretrained LLMs Learn Multiple Types of Uncertainty

Large Language Models are known to capture real-world knowledge, allowing them to excel in many downstream tasks. Despite recent advances, these models are still prone to what are commonly known...

arxiv.org
2
3
21
234
Image from Omri Fahn (@OmriFahn)
2
45
7/ 🎙️ Prefer a quick listen? • English NotebookLM podcast (8 min): notebooklm.google.com/notebo… • Hebrew NotebookLM podcast: notebooklm.google.com/notebo…
Gemini Notebook | Your research and thinking partner, grounded in the information you trust

Gemini Notebook | Your research and thinking partner, grounded in the information you trust

Use the power of AI for quick summarization and note taking, Gemini Notebook is your powerful virtual research assistant rooted in information you can trust.

notebook.google.com
1
58
8/ 📄 Read the full paper: arxiv.org/abs/2505.21218 Thanks for reading! #AI #LLM #Uncertainty #Interpretability #Research
Pretrained LLMs Learn Multiple Types of Uncertainty

Pretrained LLMs Learn Multiple Types of Uncertainty

Large Language Models are known to capture real-world knowledge, allowing them to excel in many downstream tasks. Despite recent advances, these models are still prone to what are commonly known...

arxiv.org
1
1
85
``NAST: Noise Aware Speech Tokenization for Speech Language Models,'' Shoval Messica, Yossi Adi, ift.tt/TwrHMcg
5
23
1,460
🚨I’m excited to share our #INTERSPEECH2024 paper "NAST: Noise Aware Speech Tokenization for Speech Language Models” 🥳 W/ @adiyossLC Paper-arxiv.org/abs/2406.11037 Code-github.com/ShovalMessica/NAS…
Image from Shoval Messica (@ShovalMessica)
1
14
1
32
4,229
Hello Berliners, tired of hunting for the perfect apartment or sublet? As a former exchange student in Berlin, I understand the struggle! So, I decided to let AI do the hard work. AI-Gesucht.com
Image from Omri Fahn (@OmriFahn)
1
75
more replies
Our platform is really simple to use, quicker to use than the time it takes to read this post 😉 May the AI-Gesucht guide you to your dream apartment Inside of Berlin’s ring 💍 !
1
27
AI-Gesucht.com
21
Load more