iiinova — where your digital self keeps evolving

The Curse of Verbalization: How Presentation Order Constrains LLM Reasoning aclanthology.org/2026.finding... > Restructuring problems to align the order of information presentation with the order of utilization consistently improves performance. Intuitive but neat.

The Curse of Verbalization: How Presentation Order Constrains LLM Reasoning

Yue Zhou, Henry Peng Zou, Barbara Di Eugenio, Yang Zhang. Findings of the Association for Computational Linguistics: EACL 2026. 2026.

000

Mary, the Cheeseburger-Eating Vegetarian: Do LLMs Recognize Incoherence in Narratives? aclanthology.org/2026.eacl-lo... How LLMs process contradictory information in context (wrt. what they expect or know) is a big question. Contradictions in settings seem more important to traits.

Mary, the Cheeseburger-Eating Vegetarian: Do LLMs Recognize Incoherence in Narratives?

Karin De Langis, Püren Öncel, Ryan Peters, Andrew Elfenbein, Laura Kristen Allen, Andreas Schramm, Dongyeop Kang. Proceedings of the 19th Conference of the European Chapter of the Association for Comp...

100

Rethinking Hallucinations: Correctness, Consistency, and Prompt Multiplicity aclanthology.org/2026.eacl-lo... To put it plainly, hallucinations are a frustratingly poorly defined phenomenon. To mitigate it, nuance and categorization is important, which this paper does a good job of.

Rethinking Hallucinations: Correctness, Consistency, and Prompt Multiplicity

Prakhar Ganesh, Reza Shokri, Golnoosh Farnadi. Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

100

LLMs Faithfully and Iteratively Compute Answers During CoT: A Systematic Analysis With Multi-step Arithmetics aclanthology.org/2026.finding... Big fan of faithfulness work, causal interventions and even moreso in specialized scenarios (arithmetic) where pseudolabels can be derived.

LLMs Faithfully and Iteratively Compute Answers During CoT: A Systematic Analysis With Multi-step Arithmetics

Keito Kudo, Yoichi Aoki, Tatsuki Kuribayashi, Shusaku Sone, Masaya Taniguchi, Ana Brassard, Keisuke Sakaguchi, Kentaro Inui. Findings of the Association for Computational Linguistics: EACL 2026. 2026.

100

Sycophancy Hides Linearly in the Attention Heads aclanthology.org/2026.eacl-lo... Sycophancy is a concerning phenomenon which LLMs regularly exhibit. Showing where it is encoded, and also more surprisingly, showing that it is in the attention heads, is very cool.

Sycophancy Hides Linearly in the Attention Heads

Rifo Ahmad Genadi, Munachiso Samuel Nwadike, Nurdaulet Mukhituly, Tatsuya Hiraoka, Hilal AlQuabeh, Kentaro Inui. Proceedings of the 19th Conference of the European Chapter of the Association for Compu...

100

Debora Nozza @deboranozza.bsky.social · 25/03/2026

To ease my FOMO from not attending @eaclmeeting.bsky.social, I skimmed the proceedings while playing the Tangier episode of Parts Unknown. I'll do something different and shout out 5 (subjectively) interesting works of authors I'm *not* closely related to, in no specific order:🧵⬇️

140

Reposted by Martin Tutek

Thinking of applying for an #MSCA Postdoctoral Fellowship in 2026? I’m open to supervising at Bocconi! Feel free to reach out. By submitting an expression of interest to Bocconi, selected applicants will receive full proposal support. 🗓️ Deadline: April 15 👉 www.unibocconi.it/en/horizon-e...

unibocconi.it

Horizon Europe - Marie Skłodowska-Curie Actions Postdoctoral Fellowships 2026 - Bocconi University

054

Reposted by Martin Tutek

Andreas Waldis @tresiwald.bsky.social · 25/03/2026

Excited to present this work together with @dippedrusk.com at #EACL. Join us in the poster session 1 (11:30-13:00) 🔥

041

Reposted by Martin Tutek

MilaNLP Lab @milanlp.bsky.social · 25/03/2026

Excited to share that @milanlp.bsky.social will be presenting 5 new papers at #EACL2026 and workshops in Rabat 🇲🇦!

594

Reposted by Martin Tutek

Naomi Saphra @nsaphra.bsky.social · 24/03/2026

Argh this sucks. apparently they lost their funding guarantee

4163

Reposted by Martin Tutek

Gabriella Lapesa @gabriellalapesa.bsky.social · 24/03/2026

You are #EACL2026? Check out some great work from the CSS Department @gesis.org, our data Science Methods team, with great collaborators!

4124

Reposted by Martin Tutek

Maria Antoniak @mariaa.bsky.social · 24/03/2026

I’m seeing close to zero reaction/conversation about this on here. This is huge news for open research on language models, especially in the US.

37317

Martin Tutek @mtutek.bsky.social · 24/03/2026

Yup this is a massive loss. OLMo (+entire ecosystem, OLMoTrace, the NeurIPS tutorial on the LM pipeline,...) was incredibly valuable and now I feel I took it for granted all this time. Not even counting all the great research coming out of AllenAI.

070

Reposted by Martin Tutek

Mother Jones @motherjones.com · 18/03/2026

“It’s toddler AI misinformation at an industrial scale. It’s very risky for the developing brain.” Children’s media experts say AI-generated “slop" has infiltrated the internet, preying on young children and their unsuspecting caregivers.

motherjones.com

This is your kid's brain on AI slop

“It’s toddler AI misinformation at an industrial scale. It’s very risky for the developing brain.”

19508158

Martin Tutek @mtutek.bsky.social · 18/03/2026

We'll have a reproducibility track at this years' Blackbox workshop! Details are still within a slightly opaque box. We want to see if cleaning solutions that make opaque boxes 📦 transparent 🍱 work on different boxes 🎁📮🧰🥡, and with different 🧽solution-to-water🧼ratios!

062

The probabilistic mechanism can also be calculated for closed models. We find relatively similar probabilistic results compared to open models. Given the high correlation between our probabilistic and geometric results: ➡️ We could attempt to induce the geometry of closed models!

110

This correlation dissolves in an inconsistent conversations (spanning different topics). This finding aligns with adversarial strategies that employ unrelated tokens to jailbreak models (Zou et al., 2023; Qi et al., 2025)

110

We bridge two worlds: - Probabilistic: Modeling chats as Markov chains. - Geometric: Measuring the orthogonality and dynamics in the internal state. We find a high correlation between the two; the larger the probabilistic consistency - the larger the internal trap!

110