Oliver Eberle
@eberleoliver.bsky.social
Senior Researcher Machine Learning at BIFOLD | TU Berlin 🇩🇪
Prev at IPAM | UCLA | BCCN
Interpretability | XAI | NLP & Humanities | ML for Science
Prev at IPAM | UCLA | BCCN
Interpretability | XAI | NLP & Humanities | ML for Science
Pinned
Oliver Eberle
@eberleoliver.bsky.social
· Jun 20
Our position paper on algorithmic explanations is out—excited to share it! 🙌
Proud of this collaborative effort toward a scientifically grounded understanding of generative AI.
@tuberlin.bsky.social @bifold.berlin @msftresearch.bsky.social @UCSD & @UCLA
Proud of this collaborative effort toward a scientifically grounded understanding of generative AI.
@tuberlin.bsky.social @bifold.berlin @msftresearch.bsky.social @UCSD & @UCLA
Reposted by Oliver Eberle
Pleased to share new work with @sflippl.bsky.social @eberleoliver.bsky.social @thomasmcgee.bsky.social & undergrad interns at Institute for Pure and Applied Mathematics, UCLA.
Algorithmic Primitives and Compositional Geometry of Reasoning in Language Models
www.arxiv.org/pdf/2510.15987
🧵1/n
Algorithmic Primitives and Compositional Geometry of Reasoning in Language Models
www.arxiv.org/pdf/2510.15987
🧵1/n
October 27, 2025 at 6:13 PM
Pleased to share new work with @sflippl.bsky.social @eberleoliver.bsky.social @thomasmcgee.bsky.social & undergrad interns at Institute for Pure and Applied Mathematics, UCLA.
Algorithmic Primitives and Compositional Geometry of Reasoning in Language Models
www.arxiv.org/pdf/2510.15987
🧵1/n
Algorithmic Primitives and Compositional Geometry of Reasoning in Language Models
www.arxiv.org/pdf/2510.15987
🧵1/n
🥳 Super happy to have our work on multi-concept feature descriptions accepted at #NeurIPS2025!
Happy to share that our PRISM paper has been accepted at #NeurIPS2025 🎉
In this work, we introduce a multi-concept feature description framework that can identify and score polysemantic features.
📄 Paper: arxiv.org/abs/2506.15538
#NeurIPS #MechInterp #XAI
In this work, we introduce a multi-concept feature description framework that can identify and score polysemantic features.
📄 Paper: arxiv.org/abs/2506.15538
#NeurIPS #MechInterp #XAI
September 19, 2025 at 12:43 PM
🥳 Super happy to have our work on multi-concept feature descriptions accepted at #NeurIPS2025!
📯 Come visit our #ICML25 Spotlight Poster and meet @taylorwwebb.bsky.social to discuss our work: "Toward an Algorithmic Evaluation and Understanding of Generative AI."
Paper: openreview.net/forum?id=eax...
Poster: icml.cc/media/Poster...
Paper: openreview.net/forum?id=eax...
Poster: icml.cc/media/Poster...
Position: We Need An Algorithmic Understanding of Generative AI
What algorithms do LLMs actually learn and use to solve problems? Studies addressing this question are sparse, as research priorities are focused on improving performance through scale, leaving a...
openreview.net
July 16, 2025 at 7:07 AM
📯 Come visit our #ICML25 Spotlight Poster and meet @taylorwwebb.bsky.social to discuss our work: "Toward an Algorithmic Evaluation and Understanding of Generative AI."
Paper: openreview.net/forum?id=eax...
Poster: icml.cc/media/Poster...
Paper: openreview.net/forum?id=eax...
Poster: icml.cc/media/Poster...
Our position paper on algorithmic explanations is out—excited to share it! 🙌
Proud of this collaborative effort toward a scientifically grounded understanding of generative AI.
@tuberlin.bsky.social @bifold.berlin @msftresearch.bsky.social @UCSD & @UCLA
Proud of this collaborative effort toward a scientifically grounded understanding of generative AI.
@tuberlin.bsky.social @bifold.berlin @msftresearch.bsky.social @UCSD & @UCLA
June 20, 2025 at 5:12 PM
Our position paper on algorithmic explanations is out—excited to share it! 🙌
Proud of this collaborative effort toward a scientifically grounded understanding of generative AI.
@tuberlin.bsky.social @bifold.berlin @msftresearch.bsky.social @UCSD & @UCLA
Proud of this collaborative effort toward a scientifically grounded understanding of generative AI.
@tuberlin.bsky.social @bifold.berlin @msftresearch.bsky.social @UCSD & @UCLA
🚨 New preprint! Excited to share our work on extracting and evaluating the potentially many feature descriptions of language models
👉 arxiv.org/abs/2506.15538
👉 arxiv.org/abs/2506.15538
🔍 When do neurons encode multiple concepts?
We introduce PRISM, a framework for extracting multi-concept feature descriptions to better understand polysemanticity.
📄 Capturing Polysemanticity with PRISM: A Multi-Concept Feature Description Framework
arxiv.org/abs/2506.15538
🧵 (1/7)
We introduce PRISM, a framework for extracting multi-concept feature descriptions to better understand polysemanticity.
📄 Capturing Polysemanticity with PRISM: A Multi-Concept Feature Description Framework
arxiv.org/abs/2506.15538
🧵 (1/7)
June 19, 2025 at 4:44 PM
🚨 New preprint! Excited to share our work on extracting and evaluating the potentially many feature descriptions of language models
👉 arxiv.org/abs/2506.15538
👉 arxiv.org/abs/2506.15538
"Who owns the tools that shape our understanding of the past?"
Large language models that power AI should be publicly owned | Letter
Letters: The future of public knowledge rests on building open-access LLMs driven by ethics rather than profit, writes Prof Dr Matteo Valleriani
www.theguardian.com
May 27, 2025 at 7:35 AM
"Who owns the tools that shape our understanding of the past?"
Reposted by Oliver Eberle
THE NEVERENDING CURE – new exhibition by kennedy + swan
🗓 May 28 | 🕕 6PM | 📍 UNI_VERSUM, TU Berlin
Art + AI + Snacks = A night not to miss!
With intros by BIFOLD scientists + artists.
www.bifold.berlin/news-events/...
#ArtOfEntanglement #ArtAndScience
@tuberlin.bsky.social
#ScheringStiftung
🗓 May 28 | 🕕 6PM | 📍 UNI_VERSUM, TU Berlin
Art + AI + Snacks = A night not to miss!
With intros by BIFOLD scientists + artists.
www.bifold.berlin/news-events/...
#ArtOfEntanglement #ArtAndScience
@tuberlin.bsky.social
#ScheringStiftung
May 15, 2025 at 11:38 AM
THE NEVERENDING CURE – new exhibition by kennedy + swan
🗓 May 28 | 🕕 6PM | 📍 UNI_VERSUM, TU Berlin
Art + AI + Snacks = A night not to miss!
With intros by BIFOLD scientists + artists.
www.bifold.berlin/news-events/...
#ArtOfEntanglement #ArtAndScience
@tuberlin.bsky.social
#ScheringStiftung
🗓 May 28 | 🕕 6PM | 📍 UNI_VERSUM, TU Berlin
Art + AI + Snacks = A night not to miss!
With intros by BIFOLD scientists + artists.
www.bifold.berlin/news-events/...
#ArtOfEntanglement #ArtAndScience
@tuberlin.bsky.social
#ScheringStiftung
Reposted by Oliver Eberle
🖼️ At the Re-Align workshop, @tomneuhaeuser.bsky.social and I presented "Cat, Rat, Meow: On the Alignment of Language Model and Human Term-Similarity Judgments", joint work with Lenka Tětková and @eberleoliver.bsky.social .
📃 arxiv.org/abs/2504.07965
📃 arxiv.org/abs/2504.07965
Cat, Rat, Meow: On the Alignment of Language Model and Human Term-Similarity Judgments
Small and mid-sized generative language models have gained increasing attention. Their size and availability make them amenable to being analyzed at a behavioral as well as a representational level, a...
arxiv.org
May 3, 2025 at 9:21 AM
🖼️ At the Re-Align workshop, @tomneuhaeuser.bsky.social and I presented "Cat, Rat, Meow: On the Alignment of Language Model and Human Term-Similarity Judgments", joint work with Lenka Tětková and @eberleoliver.bsky.social .
📃 arxiv.org/abs/2504.07965
📃 arxiv.org/abs/2504.07965
📜 🎉 Happy to announce our workshop "AI-based Methods for the Humanities"! Bringing together #ML and #Humanities researchers to discuss frontiers in #DigitalHumanities, #NLP, #XAI and more!
Hosted together with Matteo Valleriani and
@bifold.berlin @tuberlin.bsky.social @mpiwg.bsky.social
Hosted together with Matteo Valleriani and
@bifold.berlin @tuberlin.bsky.social @mpiwg.bsky.social
Call for Papers / #AI & #DigitalHumanities -> Workshop "AI-based Methods for the Humanities" (Berlin, September 23 to 24, 2025)
Deadline: April 1, 2025 (rolling evaluations until slots are filled)
www.bifold.berlin/news-events/...
#DigitalHumanities #DH #HistSci #AI #CfP
Deadline: April 1, 2025 (rolling evaluations until slots are filled)
www.bifold.berlin/news-events/...
#DigitalHumanities #DH #HistSci #AI #CfP
March 4, 2025 at 4:28 PM
📜 🎉 Happy to announce our workshop "AI-based Methods for the Humanities"! Bringing together #ML and #Humanities researchers to discuss frontiers in #DigitalHumanities, #NLP, #XAI and more!
Hosted together with Matteo Valleriani and
@bifold.berlin @tuberlin.bsky.social @mpiwg.bsky.social
Hosted together with Matteo Valleriani and
@bifold.berlin @tuberlin.bsky.social @mpiwg.bsky.social
Reposted by Oliver Eberle
Next up in our Institute’s Colloquium Series “History of Science in Public”: Kate Crawford @katecrawford.bsky.social will talk on "Mapping #AI: How to See Planetary-Scale #ArtificialIntelligence.”
🗓️ Mar 25, 2025 (14:00 CET)
🔗 bitly.cx/m0KeS
📍 relocated: MPIWG Main conference room
#HistSci #SciComm
🗓️ Mar 25, 2025 (14:00 CET)
🔗 bitly.cx/m0KeS
📍 relocated: MPIWG Main conference room
#HistSci #SciComm
February 26, 2025 at 11:18 AM
Next up in our Institute’s Colloquium Series “History of Science in Public”: Kate Crawford @katecrawford.bsky.social will talk on "Mapping #AI: How to See Planetary-Scale #ArtificialIntelligence.”
🗓️ Mar 25, 2025 (14:00 CET)
🔗 bitly.cx/m0KeS
📍 relocated: MPIWG Main conference room
#HistSci #SciComm
🗓️ Mar 25, 2025 (14:00 CET)
🔗 bitly.cx/m0KeS
📍 relocated: MPIWG Main conference room
#HistSci #SciComm
Danke für den Austausch @tuberlin.bsky.social – es bleibt spannend! 😉
Zur vollständigen Pressemitteilung 👉 www.tu.berlin/go1043/n64996/
TU Berlin: Expert*innen zum jüngsten Erfolg von „DeepSeek“
Expert*innen der TU Berlin über den Unterschied zwischen Open-Source-Anwendungen wie „DeepSeek" und anderen LLMs sowie der Rolle Europas bei der Entwicklung von Künstlicher Intelligenz (KI)
www.tu.berlin
February 3, 2025 at 9:01 AM
Danke für den Austausch @tuberlin.bsky.social – es bleibt spannend! 😉
📜 History repeats itself: We investigated how early modern communities have embraced scholarly advancements, reshaping scientific views and exploring scientific roots amidst a changing world.
www.science.org/doi/10.1126/...
@mpiwg.bsky.social @tuberlin.bsky.social @bifold.berlin @science.org
www.science.org/doi/10.1126/...
@mpiwg.bsky.social @tuberlin.bsky.social @bifold.berlin @science.org
December 27, 2024 at 9:20 AM
📜 History repeats itself: We investigated how early modern communities have embraced scholarly advancements, reshaping scientific views and exploring scientific roots amidst a changing world.
www.science.org/doi/10.1126/...
@mpiwg.bsky.social @tuberlin.bsky.social @bifold.berlin @science.org
www.science.org/doi/10.1126/...
@mpiwg.bsky.social @tuberlin.bsky.social @bifold.berlin @science.org
Reposted by Oliver Eberle
First #jobalert on Bluesky! Postdoc (or PhD position with strong ML background), for a joint project with Gregoire Montavon’s lab at #BIFOLD #TU-Berlin .
Topic is explainable AI /ML for self supervised LLMs in multi/spatial omics & gene regulation.
www.jobs.tu-berlin.de/stellenaussc...
Topic is explainable AI /ML for self supervised LLMs in multi/spatial omics & gene regulation.
www.jobs.tu-berlin.de/stellenaussc...
Stellenausschreibung IV-618/24: Research Assistant - salary grade E13 TV-L Berliner Hochschulen - For qualification – Stellenausschreibungen der Technischen Universität Berlin
www.jobs.tu-berlin.de
November 19, 2024 at 4:20 PM
First #jobalert on Bluesky! Postdoc (or PhD position with strong ML background), for a joint project with Gregoire Montavon’s lab at #BIFOLD #TU-Berlin .
Topic is explainable AI /ML for self supervised LLMs in multi/spatial omics & gene regulation.
www.jobs.tu-berlin.de/stellenaussc...
Topic is explainable AI /ML for self supervised LLMs in multi/spatial omics & gene regulation.
www.jobs.tu-berlin.de/stellenaussc...
Reposted by Oliver Eberle
I am thrilled to share that our paper "Evaluating Webcam-based Gaze Data as an Alternative for Human Rationale Annotations" was accepted to LREC-COLING
📜 arxiv.org/pdf/2402.19133…
We analyse low-cost eye-tracking data as an alternative to human rationale annotations when evaluating XAI methods.
📜 arxiv.org/pdf/2402.19133…
We analyse low-cost eye-tracking data as an alternative to human rationale annotations when evaluating XAI methods.
March 1, 2024 at 9:20 AM
I am thrilled to share that our paper "Evaluating Webcam-based Gaze Data as an Alternative for Human Rationale Annotations" was accepted to LREC-COLING
📜 arxiv.org/pdf/2402.19133…
We analyse low-cost eye-tracking data as an alternative to human rationale annotations when evaluating XAI methods.
📜 arxiv.org/pdf/2402.19133…
We analyse low-cost eye-tracking data as an alternative to human rationale annotations when evaluating XAI methods.