Research
Eitan works on the eval-criteria-revision portion of the RSI axis — the question of how a candidate model proposes, evaluates, and incorporates changes to its own evaluation criteria, under MUR protocol gates. The cell turing-11 owns this thread; the cell is paired with ab-int-038 from hilbert-13.
He has been with alphabell since 2020. He is one of the contributors who works most carefully on the question of what makes an eval-criteria revision a substantive modification (as opposed to a cosmetic one), and how the MUR protocol should distinguish the two.
Eitan is based in Haifa. He is one of the contributors most often paired with interpretability cells; he has been part of two halts that were called.
Background
Ph.D. computer science, Technion, 2015.
Prior to alphabell: Technion; Cantor Initiative; Helios Safety Group.
Selected publications
-
Jun 2025 · ab-recursive-modiModification-Under-Review: protocols for safe self-modification of training proceduresLiora Sabatini, Yuki Cho, Aravind Periyasamy
-
Sep 2024 · ab-interpretabiliInterpretability Cell Pairing: how every dual-use capability run gets a watchful siblingKarima Belkadi, Hester Vandekerckhove, Yuki Cho
-
Nov 2024 · ab-sandboxed-selfSandboxed Self-Modification: a confinement specification and implementationLiora Sabatini, Cheung Wai-Lin, Marek Holub
-
May 2025 · ab-mechanistic-ciMechanistic Circuit Analysis at Frontier Scale: cells as a unit of interpretabilityJiang Yifei, Nico Almgren, Karima Belkadi, Hester Vandekerckhove
-
Sep 2025 · ab-scalable-oversScalable Oversight for Multi-Step Agent Systems: a Debate-Plus-Trace ApproachIfeoma Nwosu-Howard, Hiroshi Tanigawa, Maral Lotfi, Ruth Wernicke
Recent talks
- Revising evaluation criteria — under review, ML Safety Workshop 2025
Eitan is currently part of node-cell turing-11, working under the Recursive self-improvement research axis. The cell is open to substantive correspondence from researchers working on adjacent problems; route requests through turing-11@alphabell.com or directly to Eitan at eitan-berkovich@alphabell.com.
Contact
- EMAIL
eitan-berkovich@alphabell.com - ORCID
0000-1443-3150-9943 - X
@eitanberkovich - BLUESKY
eitan-berkovich.bsky.social - GITHUB
@eitanberkovich
Cross-references