Pascal Tikeng (migrant)

Learning something ... at @Mila_Quebec

[Tara Research @TaraResearch]

🚨✨Today we're introducing Tara Research, and sharing our first public work: a paper accepted at COLM 2026 (with a companion LessWrong post) ��.

Who we are: Tara Research is an independent, non-profit AI safety organization. We measure the propensity of AI systems to lie, and we benchmark how well safety techniques prevent it, openly and impartially. We believe that misalignment becomes catastrophic when models hide it. A model that is honest about its actions and goals can be corrected; one that lies cannot.

The paper: "Activation Steering for Aligned Open-ended Generation without Sacrificing Coherence."

We introduce two projection-aware steering methods that correct only the tokens whose activations fall on the misaligned side of a learned decision boundary. They restore honesty as well as classic uniform steering, at a fraction of the capability cost.

Additionally, we find a single honesty direction, extracted with our contrastive dataset, that generalizes and corrects dishonesty across four unrelated settings:
- Instructed lying (MASK benchmark): honesty 53.6% → 81.2%
- Strategic deception (Among Us): crewmate win rate 49% → 95%
- Hidden trained behaviors (AuditBench): discovery rate ~40% → ~85%
- Emergent misalignment: honesty 46.2% → 79.4%

In the last two settings, the vector was extracted from the aligned checkpoint and applied to a model that was subsequently fine-tuned, and it still worked.

This work is by Niklas Herbster, Martin Zborowski, Alberto Tosato, @gauthier_gidel, and @Tommaso_Tosato, in collaboration with @Mila_Quebec and the @TU_Muenchen .

🌐 About us: https://t.co/wMgL0BOjVA

📄 Paper: https://t.co/O2IkWJC1Ps
📝 Blog post: https://t.co/TOc0ZmKE4e

This is the first of much more to come, including an agentic honesty benchmark. Follow us to stay informed. And if you work on runtime interventions, honesty evaluation, or alignment auditing, we'd love to hear from you.

[Marawan Gamal @mrremila]

Excited to announce that I will be presenting Model Merging via Data-Free Covariance Estimation at COLM '26!

Thanks again to my amazing co-authors @dtredsox13 @PTikeng Colin Raffel and Guillaume Rabusseau

[Marawan Gamal @mrremila]

Excited to be giving an Oral at the @icmlconf Continual Adaptation at Scale workshop today at 1:50 PM.

I'll be talking about our recent work on obtaining 𝗱𝗮𝘁𝗮-𝗳𝗿𝗲𝗲 𝗰𝗼𝘃𝗮𝗿𝗶𝗮𝗻𝗰𝗲 𝗲𝘀𝘁𝗶𝗺𝗮𝘁𝗲𝘀 of activations, for model merging.

[Marawan Gamal @mrremila]

Excited to share that our work on Model Merging has been accepted for an Oral Presentation at the CATS Workshop @ ICML 2026!

Looking forward to presenting our work next week and connecting with everyone at the workshop.

#ICML2026
https://t.co/LQmAcrghOJ

[Pascal Tikeng (migrant) @PTikeng]

Happy to be recognized as an ICML Gold Reviewer this year.

[ICML Conference @icmlconf]

Reviewers & ACs for #ICML2026 have been recognized for their service!

- Reviewers: 4439 Gold (free registration), 4437 Silver. 17749 total reviewers were assigned >= 1 paper
- ACs: 1647 receive free registration, out of 1691 who were assigned >= 1 paper

TY for your hard work!

[Marawan Gamal @mrremila]

If you want to semantically search through ICLR papers this week, check out https://t.co/4hlcIejLmL

[Marawan Gamal @mrremila]

New preprint! Introducing ACTMat: Model Merging via Data-Free Covariance Estimation

Work done w/ @dtredsox13, @PTikeng, Colin Raffel and Guillaume Rabusseau

TL;DR: It's RegMean, but without needing data for covariance estimation (C ≈ Δᵀ Δ)

📄 https://t.co/qB0oafSxvq
🧵(1/6)

[Pascal Tikeng (migrant) @PTikeng]

@KempeLab How to provide feedbacks? ("Openreview" style)

[Suffiyan Malik @suffiyanmalikk]

@dwarkesh_sp 9pm-3am is also incredible focus time

[Pascal Tikeng (migrant) @PTikeng]

@introspection @Mila_Quebec 2) We also show that the commonly used L2 norm is not a reliable proxy for explaining grokking.

3) Finally, we investigate how factors such as training dataset choice and model overparameterization impact grokking delay

The figure below summarizes our contribution fairly well

[Pascal Tikeng (migrant) @PTikeng]

@introspection @Mila_Quebec 1) We show that grokking time scales proportionally to 1/(α * β) when minimizing composite objectives of the form f = g + βh using gradient descent with learning rate α, where g is the training error and h is any regularizer that enforces an inductive bias toward generalization.

[Pascal Tikeng (migrant) @PTikeng]

@ferezola @tassingremi @arol_ketch J’avais acheté le mien ici https://t.co/CcLBJ8X0VC

[Guillaume Dumas @introspection]

📝New preprint out!🔍 We used dynamical systems theory to predict "Grokking" — the phase transition leading to "Generalization Beyond Overfitting" discovered at @OpenAI last year— Big kudos to our @Mila_Quebec dream team, especially @PTikeng!