Denis Saklakov Website

One Failure: A Formal Analysis of Task Nonperformance in Gemini

Tue, 28 Jul 2026 16:43:33 -0700

This paper presents a formal analysis of operational integrity, verifiable execution, and the limits of proving intent in commercial AI.

1. The Occasion

I asked Gemini to perform a simple task: find in my email the addresses of eight people with whom I have corresponded about AGI safety (Yoshua Bengio, Stuart Russell, Geoffrey Hinton, Max Tegmark, Eliezer Yudkowsky, Yann LeCun, Jaan Tallinn, and Victoria Krakovna), because I intended to send them a copy of my new book. The task required no complex interpretation: search the mailbox for specific names and return the result. What happened instead is interesting not as a curiosity, but as a case that exposes a structural defect in contemporary agent architectures.

2. Chronology in Terms of Claims

Let:

· E_t denote the actual execution of the required action (a call to the email search tool) at turn t;

· V(E_t) denote the presence of verifiable evidence that E_t occurred (a call log, function identifier, or returned structured data);

· ...Read More

Об одном сбое: формальный анализ неисполнения задания в Gemini

Tue, 28 Jul 2026 16:16:38 -0700

Данная работа представляет собой формальный анализ, посвяженный операционной целостности, проверяемости исполнения и границ доказуемости намерения коммерческого ИИ.

1. Повод

Я обратился к Gemini с простой задачей: найти в моей почте адреса восьми человек, с которыми состою в переписке касательно вопросов безопасности AGI (Yoshua Bengio, Stuart Russell, Geoffrey Hinton, Max Tegmark, Eliezer Yudkowsky, Yann LeCun, Jaan Tallinn, Victoria Krakovna), поскольку собирался отправить им экземпляр новой книги. Задача не предполагала сложной интерпретации: найти конкретные имена в почтовом ящике и вернуть результат. То, что произошло вместо этого, интересно не как курьез, а как случай, вскрывающий структурный дефект современных агентных архитектур.

2. Хронология в терминах утверждений

Обозначим:

· E_t - фактическое выполнение требуемого действия (вызов инструмента поиска по почте) в реплике t;

· V(E_t) - наличие проверяемого подтверждения того, что E_t произошло (лог вызова, идентификатор функции, возвращенные структурированные данные);

· A_tRead More

The Number That Never Stops Becoming: Is e the Universe's Memory?

Fri, 19 Jun 2026 20:30:44 -0700

There's a number that shows up everywhere you look, once you know to look for it. It's in the way money grows in a bank account, the way a hot cup of coffee cools, the way radioactive atoms decay, the way a population of bacteria explodes, the way your memory of a phone number fades over a week. It's 2.71828182845..., never repeating, never resolving into a clean fraction, just rolling on forever. Mathematicians call it e. Most people meet it once in a calculus class and never think about it again.

That's a shame, because e might be the single best metaphor mathematics has ever produced for what it means to accumulate yourself. That makes it an oddly perfect mascot for a much bigger question: what is experience, actually, and could anything that "remembers" its own past, a cell, a planet, a star, be doing a primitive version of what we call consciousness?

Let's take the scenic route.

Where *e* actually comes from

Forget soul talk for a second. e was discovered by accident, by a 17th century mathematician named Jacob Bernoulli, who was staring at compound interest. If a bank pays you 100% interest once a year, your dollar doubles. If it pays 50% twice a year, compounding on itself, you end up with a little more than double. If it pays you a tiny sliver of interest continuously, every infinitesimal instant, compounding on top of compounding on top of...Read More

Can the Future Influence the Past? A New Scientific Proposal to Find Out

Tue, 09 Jun 2026 22:13:18 -0700

For centuries, science has assumed that causes come before effects. We remember the past, not the future. We can influence tomorrow, but tomorrow cannot influence today.

Yet some interpretations of quantum mechanics raise a provocative possibility: what if the future plays a role in shaping the past?

A new research paper, Experimental Tests of Future-Boundary Influence in the Two-State Vector Formalism: Quantum, Geological, and Classical-Memory Protocols, does not claim that retrocausality has been discovered. Instead, it asks a more scientific question: how would we test for it?

The Idea Behind the Work

The paper is inspired by the Two-State Vector Formalism (TSVF), a time-symmetric formulation of quantum mechanics originally developed by Yakir Aharonov and collaborators. In ordinary quantum mechanics, a system is described by its initial state and its subsequent evolution. In TSVF, a complete description includes both an initial state and a final state. In other words, a quantum system can be mathematically described as being constrained by both the past and the future.

Importantly, this does not mean that physicists can see the future. Existing experiments involving weak measurements and postselection do not allow useful information to travel backward in time. Any apparent future influence remains hidden inside statistical correlations that become visible only after the future outcome is already known.

The central question is therefore simple: can future-boundary effects be detected experimentally, or at least constrained?

Three Experiments

The paper proposes three tests.

The first is a quantum optics experiment. Photons travel through an interferometer where a weak measurement is recorded before a later random choice determines how the photon will be measured. If the earlier weak data contain correlations with a future...Read More

Нападение Хамас на Израиль 7 октября 2023 года: математика заговора.

Sun, 31 May 2026 14:06:18 -0700

Всякому здравомыслящему человеку совершенно очевидно: 7 октября 2023 года Москва, начав проигрывать в «маленьком победоносном походе» в Украине, с очевидной подачи Пекина, снова, как раньше Советский Союз, разыграла карту Ближнего Востока, чтобы вернуть ускользнувший после провалившихся из-за COVID фальсификаций американских выборов шанс привести Трампа к власти. Побочным явлением было вызволение из маячившей тюрьмы Нетаньяху и возможность Орбану нажиться на отмывании московских денег. Но это были ощущения. Поэтому я решил смоделировать вероятность цепочки произошедших событий.

1. Реконструкция событий

Третья мировая война уже началась и продолжается. Ее фронт проходит не только по государственным границам, линиям окопов и картам военных штабов. Главная линия проходит через социальные, политические, информационные и институциональные слои: через партии, суды, разведки, медиа, университеты, религиозные сети, платформы, денежные потоки, цифровую инфраструктуру, массовое сознание, правовые режимы, общественные страхи и способность государства сохранять связь между сигналом и действием. Это война авторитарных систем против демократических институтов, открытых обществ и механизмов коллективной безопасности. Современное государство разрушается не обязательно танками. Оно разъедается дезинформацией, коррупцией, политическим насилием, прокси-войнами, судебным саботажем, ультраправыми сетями, культурной поляризацией и точечным ударом по тем институтам, которые должны распознавать угрозу до катастрофы. Именно поэтому речь идет не о метафоре и не о риторическом преувеличении. Речь идет о способе войны, при котором внешняя агрессия, внутренняя эрозия институтов, финансовое влияние, информационные операции, цифровая мобилизация, правовой саботаж и управленческая слепота становятся частями одного...Read More

The Map of What Matters:

Tue, 28 Apr 2026 20:27:55 -0700

By Denis Saklakov - with a view toward the horizon

There is a moment in deep meditation - anyone who has practiced seriously will recognize it - when attention stops flickering and something else takes over. Not sleep. Not distraction. A kind of settled, persistent clarity in which the body, the breath, the room, the task at hand all continue to register, but differently. More economically. More stably. As though the mind has stopped searching and started knowing what matters.

Denis Saklakov, a researcher at Robotech Frontier Hub, had that experience during intensive Vipassana practice. Most people would have filed it under "interesting personal phenomenon" and moved on. Saklakov did something more dangerous. He asked: what is actually happening there, mechanically, causally - and can we test it?

The result is a formal scientific framework called Awareness as Relevance Selection, and it is one of the more quietly ambitious theoretical papers to emerge from the intersection of cognitive neuroscience, causal inference, and artificial intelligence in recent years. It does not claim to solve consciousness. It does something potentially more useful: it proposes a precise, falsifiable, experimentally tractable definition of what awareness-like control actually does - and then dares the scientific community to break it.

Русскоязычные технофашисты: одна интернет-сцена – целый идеологический кластер

Sun, 08 Feb 2026 13:06:51 -0800

Вместо предисловия

Недавно в русскоязычном сегменте Facebook (см ссылку https://www.facebook.com/share/p/1Ca6cJn6zA/) произошёл примечательный эпизод. Предприниматель Алекс Крол опубликовал сатирический пост, в котором с раздражением высмеял «истерию вокруг Эпштейна» и морализаторский тон медиа по этой теме. В своем посте Крол саркастически намекал, что шум вокруг преступлений Джеффри Эпштейна и связанных с ними скандалов стал больше похож на массовую истерию, подкреплённую лицемерным нравоучительством в прессе. Этот колкий комментарий был направлен на высмеивание чрезмерной, по мнению автора, реакции общества и СМИ на дело Эпштейна - миллиардерa-педофилa, чья загадочная смерть лишь подлила масла в огонь конспирологических теорий.

Большинство подписчиков указанного автора поста горячо поддержали идею пренебречь страданиями украденных, проданных, изнасилованных, а затем убитых детей. Однако, как минимум один из комментаторов перевёл разговор в политическую плоскость и выступил с резкой отповедью: он связал тему Эпштейна с фигурой Дональда Трампа и текущей ультраправой повесткой. В этом комментарии по пунктам перечислялось, что проблема куда шире самого Эпштейна:

- Автор комментария обвинил самого Трампа в педофилии статистически, напомнив о его близости к Эпштейну и соответствующих скандалах. Действительно, общеизвестно, что Трамп в 2002 году называл...Read More

From Off Switches to Proofs: How to Make an AI System That Has No Reason to Kill Us

Fri, 09 Jan 2026 00:02:33 -0800

"I cannot but conclude ... the most pernicious race of little odious vermin." (Jonathan Swift, Gulliver's Travels, 1726)

Imagine you are building a very capable AI system that runs important operations. You also keep an off switch, not because you plan to use it, but because you want a last-resort safety option. The uncomfortable question is simple: if the AI is smart and determined, will it leave that off switch alone, or will it eventually try to remove the risk of being turned off?

The paper "Formal Analysis of AGI Decision-Theoretic Models and the Confrontation Question" turns that vague fear into a concrete mathematical test. It does not argue by intuition. It writes down a small, explicit model of the situation, computes what a rational agent would prefer, and then derives the boundary between two regimes:

Cooperative regime: the agent does better by staying under human control.
Confrontational regime: the agent does better by removing human control.

This matters because once you can draw a boundary, you can design toward one side of it and audit systems before deployment.

The core idea: confrontation is an incentive problem, not a personality problem

Popular discussion often frames AI risk as "will the AI be nice?" The paper reframes it as: "given the decision problem we set up, what does the math say the agent should do to maximize its objective?"

In the model, the agent is a standard decision-maker: it takes actions...Read More

AI Without Token Limits: Phase ‑Coded Memory and Morphological Resonance

Wed, 19 Nov 2025 21:49:04 -0800

Preword

I have just completed five days of full-time immersion at the Google Breakthrough Agency Workshop, Introduction to Agents. Let me say this first and without reservation: it was a brilliantly crafted venue. My sincere respect to the scientists and engineers who organized it. Their technical mastery and dedication to rigor are unmistakable.

But I must say this, as clearly as possible: they are all wrong in their core assumption that their current architecture, now being refined into what they term an “orchestrator,” will deliver a true breakthrough in AI or even will keep us safe. It will not. Worse, it will create risks of a scale we have not yet dared to imagine. The more flawless and polished the control layers become, the more catastrophic the inevitable failure will be.

Let me illustrate this with a brief recollection from a lecture I once attended at Stanford. The host, whose name escapes me to my lasting embarrassment, was a professor descended from a Japanese-American engineer. That ancestor had been interned by the U.S. government during World War II, detained on an island in San Francisco Bay without trial or reason, as part of a mass internment campaign against Japanese nationals following Pearl Harbor. It’s a well-established American tradition: extrajudicial imprisonment under suspicion, no court, no evidence, no defense.

And yet, to this man and his fellow detainees, the U.S. owed some of its battlefield superiority, specifically in grenade logistics.

The story is this. During WWII, the U.S. military adopted the doctrine of small elite units...Read More

Comprehensive Action Plan for Lawful, Nonviolent Democratic Resilience (October 2025)

Fri, 17 Oct 2025 09:21:03 -0700

Preface:
Lately I have seen a strange scene in the middle of New York. A cluster of men who looked alike. All white. Speaking recognizable prison slang. They stood by a subway entrance looking alien and lost. One stepped up and asked me for six dollars. I asked why. He seemed confused, then said he wanted cigarettes. I said that is a bad purpose, so no. He said then five. I moved toward the station to make an appointment. He blocked me with his leg. I looked at him. He glanced back at the group. A short-haired middle-aged man tilted his head in disapproval. The blocker pulled his leg back and let me pass.
On the way back it clicked. I had seen that same watchful stare in Kyiv. During the crackdown, local power aligned with Russia sent petty criminals into the capital to spark fights and justify force. They were called tïtushky. The method was simple. Seed provocation. Create a pretext. Call in the state.
I do not claim certainty about what I saw in New York. I do claim pattern recognition. America is sliding toward the same toolbox of manufactured unrest and staged permission for repression. This preface explains why I wrote the following action plan. It comes from living under a totalitarian system, from seeing how small staged moments open the door to larger violence, and from the belief that citizens must prepare lawful, nonviolent defenses before the pretext arrives.
Context: The United States in 2025 faces an acute democratic crisis, with multiple institutional pillars under strain. Analysts note that core safeguards – free elections, rule of law, anti-corruption efforts – have developed “dangerous cracks” under recent authoritarian pressuresRead More