Research, in the open.
Everything the lab publishes, in full. Confirmations and failures alike — a result that did not replicate is part of the record, not an omission from it.
- Replication reports
- 0
- Essays
- 5
5 papers matching these filters
Replication reports
0 papersNo replication reports match this view.
- August 29, 2026Revised August 29, 202614 min readEssayComputeAIMethods
The Recursive Self-Improvement Loop: A Specification, Not a Prophecy
Recursive self-improvement (RSI) — the process by which an intelligent system improves its own capacity to improve itself — is usually discussed in one of two failure modes: as an inevitability (the "intelligence explosion" reading) or as an impossibility (the "diminishing return…
Oleksii Vergunenko - August 29, 2026Revised September 16, 202617 min readEssayComputePlatformAIMethods
Aligning Artificial Intelligence with Humanity: Problem Structure, Current Approaches, and Open Problems
The alignment problem asks whether advanced artificial intelligence systems can be made to reliably pursue objectives compatible with human values and intentions, and to continue doing so as their capabilities scale beyond the capacity of humans to directly supervise them. This p…
Oleksii Vergunenko - August 29, 2026Revised September 16, 202612 min readEssayComputeAIMethods
Alignment as Specification: A Survey of Technical AI Safety and Its Open Problems
As machine learning systems approach and exceed human performance on economically significant tasks, the question of whether their behavior can be reliably specified, verified, and controlled has moved from philosophical speculation to an active engineering discipline. This paper…
Oleksii Vergunenko - August 29, 2026Revised September 16, 202626 min readEssayComputeBioTechPlatformAIMethods
The Verification Oracle Determines the Ceiling: Why Artificial Intelligence Advances Unevenly Across Mathematics, Physics, and Biology
Between mid-2025 and mid-2026, artificial intelligence systems moved from competition-level performance to contributions at the research frontier in mathematics, while in physics and biology the same underlying models were deployed largely as surrogates, controllers, and hypothes…
Oleksii Vergunenko - August 29, 2026Revised August 30, 202619 min readEssayComputeAIPhysicsMaterialsEngineering
Multiplication Without Transistors: An Audit of Photonic Matrix Processors as Accelerators for Large Language Model Training
Matrix multiplication accounts for the overwhelming majority of arithmetic operations — and a dominant share of energy expenditure — in training large language models (LLMs). Photonic processors perform matrix–vector multiplication (MVM) as a passive consequence of light propaga…
Oleksii Vergunenko