Lifeboat Foundation: Safeguarding Humanity
Toggle light / dark theme

Blog

Category: robotics/AI

Resolving Feynman’s restaurant problem reveals optimal solutions and human strategies

They reconstructed Feynman’s “restaurant problem” and proved that his solution was mathematically optimal. The challenge belongs to a class of problems known as “explore versus exploit” decisions—a tradeoff that appears everywhere from choosing restaurants and dating partners to scientific research and artificial intelligence. Explore too much, and you waste opportunities enjoying known good options. Exploit too soon, and you might miss something even better.

Feynman’s restaurant problem is an instance of what is known as an optimal stopping problem (7, 8). As such, it falls in the same category as the famous secretary problem (9), in which an interviewer seeks to maximize the probability of hiring the best candidate for a position but can only evaluate those candidates relative to one another. This problem can be translated to the dining setting by assuming the goal is to maximize the probability of selecting the best restaurant over a series of meals. However, Feynman’s problem differs from the classic secretary problem in three ways: the distribution from which the restaurants are drawn is known, the diner is able to return to restaurants that they visited previously, and the goal is to maximize the total score across nights rather than the probability of identifying the single best option.

Feynman’s restaurant problem is also closely related to the finite-horizon multi-armed bandit problem (10, 11), in which a decision-maker is presented with a set of options that differ in their payoffs (such as different arms of a gambling machine) and seeks to maximize the total payoff received from trying those options a fixed number of times. Again, this could be translated to the dining context, treating the restaurants as the different options. The key difference here is that in the multiarmed bandit problem the payoffs are usually stochastic, with a distribution around the true value, while in Feynman’s problem the true value of a restaurant is directly observed. Like the multiarmed bandit problem, Feynman’s restaurant problem creates a tension between exploring new options and exploiting knowledge acquired so far, but does so without dealing with uncertain observations.

Optimal stopping problems often arise in everyday life, appearing not just in choosing what to eat, but in finding a home, deciding who to marry, selecting a parking spot, and knowing when to quit a job (12). Extensive literatures have explored how people solve variants of the secretary problem (13 – 17) and the multiarmed bandit problem (18 – 25). Feynman’s restaurant problem is a valuable addition to this canon: by removing uncertain observations, it makes it possible to study the explore–exploit tradeoff in a particularly pure manner, and the existence of closed-form expressions for the optimal policy facilitates its comparison to human behavior. In fact, previous behavioral experiments have used optimal stopping problems that are similar to (26 – 28) or variants on (29 and 30) Feynman’s restaurant problem (for a detailed breakdown see Discussion). Here, we make use of the optimal solutions we derived for multiple variants of this problem, together with an innovative experimental design that allows us to get an unusually clear picture of people’s behavior and to draw direct parallels between results in the psychological literature and the solution found by Feynman. As a consequence, we hope to not just re-solve the problem that Feynman first posed more than 40 y ago, but to resolve the question of how people perform such tasks.

Are We the Bootloader for Superintelligence?

A 90 minute interview about AI and our human future.

Dr. Hugo de Garis is a computer scientist, AI researcher, and former professor known for his early work on evolvable hardware, artificial brains, and the long-term risks of superintelligent machines. He coined and popularized the idea of the “Artilect War,” a future conflict between those who want to build godlike artificial intellects and those who believe such systems pose an existential threat to humanity. In the interview, he describes himself as trained in pure mathematics and theoretical physics, formerly a computer science professor, and now focused on broader questions about AI, cosmology, civilization, and the future of humanity.

The interview with Prof. Hugo de Garis centers on his long-standing warning that humanity may face an “Artilect War,” a civilizational conflict over whether to build godlike artificial intellects vastly superior to humans. De Garis argues that future computation, potentially extending from nanotech to femtotech and beyond, could produce minds trillions of trillions of times more capable than ours. He distinguishes between Cosmists, who want to build such beings to expand intelligence into the universe, and Terrans, who oppose them because superintelligence may eliminate or marginalize humanity. He personally remains torn, admiring the cosmic grandeur of posthuman intelligence while recognizing the existential danger.

The conversation also covers AI timelines, recursive self-improvement, AI alignment, the U.S.-China race, the Fermi paradox, simulation theory, cyborgs, cryonics, AI-generated content, the decline of universities, and the future of work. De Garis is impressed by current AI systems, treating them almost as intellectual companions, but he doubts that humanity can guarantee long-term control over recursively improving machines. The central theme is that the question “Should humanity build artilects?” may become the defining political and moral problem of the twenty-first century.

Website https://profhugodegaris.wordpress.com… is Roman Yampolskiy: https://grokipedia.com/page/roman_yam… Research papers: https://scholar.google.com/citations?… Books: AI: Unexplainable, Unpredictable, Uncontrollable https://www.amazon.com/Unexplainable-?tag=lifeboatfound-20… Considerations on the AI Endgame https://www.amazon.com/Considerations?tag=lifeboatfound-20… Artificial Superintelligence: A Futuristic Approach https://www.amazon.com/Artificial-Sup?tag=lifeboatfound-20… Artificial Intelligence Safety and Security https://www.amazon.com/Artificial-Int?tag=lifeboatfound-20… Social Media X https://twitter.com/romanyam FB / roman.yampolskiy IN / romanyam Ask Roman to speak at your event: https://www.romanyampolskiy.com/

Continue reading “Are We the Bootloader for Superintelligence?” | >

Claude Fable 5: Why This Time is Different

Anthropic has released Claude Fable 5, the public version of a model it once said was too dangerous to ship.

Ethan Mollick spent years describing AI use as casting spells, and now he’s not sure he’s the wizard anymore.

The model ran for nine and a half hours unsupervised and delivered working software he could barely fault.

But the same system hands its most dangerous capabilities to the NSA while the public gets the sanded-down version.

This video asks what it means when the most useful AI is also the least legible.

Sources to Google.

Continue reading “Claude Fable 5: Why This Time is Different” | >

AI Agent Benchmark for Real-World Professional Workflows

To solve this “utility problem,” researchers have introduced a rigorous new testing ground called Agents’ Last Exam (ALE). The name carries a dual meaning: it acts as a final graduation exam to prove an AI agent is actually ready for corporate deployment, and it represents the absolute frontier of what today’s technology can handle.

The creators of ALE don’t intend for it to be a static, one-time leaderboard. Designed as a “living benchmark,” its pool of tests will continuously grow as new industries and workflows evolve. Ultimately, the goal of Agents’ Last Exam is to shift the AI industry’s focus away from winning abstract academic trophies and toward creating digital assistants capable of driving genuine, measurable economic growth.

Challenge and measure AI agents on economically valuable and real-world tasks.

Agents’ Last Exam is building the largest-scale, broadest-coverage agent evaluation benchmark to date, measuring performance on long-horizon, economically valuable tasks with verifiable outcomes. Led by Berkeley RDI and 300+ industry experts, it now spans all 55 targeted sub-industries covering most major fields of professional work performed on a computer, with 1,500+ tasks collected toward a 5,000-task target, keeping scores objective, comparable, and meaningful across domains.

AI is incapable of telling the truth

We worry that AI will spread misinformation, but the real problem runs deeper: AI is incapable of telling the truth at all. Philosophers Bun-Sun Kim and Hongjoon Jo draw on Foucault and Heidegger to argue that humans speak truthfully because our finite, mortal existence is at stake in every word we say. AI, lacking a body, anxiety, or a conscience, risks nothing — it just recombines the internet’s idle talk into statistically plausible text, with no self to reveal. Outsourcing our communication to AI doesn’t just degrade information; it traps us in an endless loop of crowd-sourced mimicry, and threatens our capacity for genuine thought.

ChatGPT can answer complex questions and even seem to hold conversations. But can it tell the truth?

In an era where AI can answer virtually any human question, we must examine whether AI language can truly contain truth. Since the Dartmouth Conference of 1956, we’ve witnessed dramatic technological evolution—from the AI Winter of the 1970s and 80s to today’s sophisticated language models like ChatGPT that generate remarkably human-like text. As we increasingly delegate communication to artificial, rather than human, entities, a fundamental question emerges: Can AI’s artificial language capture the essence of truth conveyed by human discourse?

Kyocera develops breakthrough multilayer ceramic core substrate for advanced AI semiconductors

face_with_colon_three I still think that ceramics would be very useful to stop the need for global mining operations that rely heavily on rare materials when they can make the same chip from ceramics.

To be shown at ECTC 2026, May 26–29 in Orlando, USA, the new substrate technology delivers superior rigidity and circuit miniaturization for next-gen data centers, AI, and ASIC packaging.

Future AI chips could be built on glass

The idea is to use glass as the substrate, or layer, on which multiple silicon chips are connected. This form of “packaging” is an increasingly popular way to build computing hardware, because it lets engineers combine specialized chips designed for specific functions into a single system. But it presents challenges, including the fact that hardworking chips can run so hot they physically warp the substrate they’re built on. This can lead to misaligned components and may reduce how efficiently the chips can be cooled, leading to damage or premature failure.

“As AI workloads surge and package sizes expand, the industry is confronting very real mechanical constraints that impact the trajectory of high-performance computing,” says Deepak Kulkarni, a senior fellow at the chip design company Advanced Micro Devices (AMD). “One of the most fundamental is warpage.”

That’s where glass comes in. It can handle the added heat better than existing substrates, and it will let engineers keep shrinking chip packages—which will make them faster and more energy efficient. It “unlocks the ability to keep scaling package footprints without hitting a mechanical wall,” says Kulkarni.

Taking Longer Steps in Numerical Simulations

It’s often the case that a dynamical system’s constituents move orders of magnitude more quickly than the collective motion that interests researchers. That disparity in scale frustrates modelers. So many computationally intensive time steps are needed to reach the final state that the computation becomes infeasible. Now Filippo Bigi of the Swiss Federal Institute of Technology in Lausanne (EPFL) and his colleagues have extended and tested an approach that uses a machine-learning model to extend the time steps in an atomic-scale simulation by an order of magnitude or more while obeying physical constraints [1]. Their method is general and could be applied to planetary systems, molecular machines, and other dynamical systems.

The EPFL researchers’ starting point was a formulation of classical mechanics that describes the evolution of a system in terms of the positions and momenta of its constituents and an energy term, the Hamiltonian. In general, these and other equations of classical mechanics satisfy fundamental geometric constraints. What’s more, approximate solutions of those equations can be made to satisfy the same constraints. Bigi and his colleagues realized that machine learning could leapfrog over many time steps while also respecting those same geometric constraints.

The researchers tested their approach on several systems, including the three-body problem of celestial dynamics and the transition of germanium telluride to a glassy state. Their simulations reproduced trusted benchmarks but with time steps ten or so times longer. Currently, enforcing the physical constraints undoes most of the computational advantage of the longer time steps. However, the team is optimistic that it can find more computationally efficient implementations.

To discover new physics, AI may need to ‘unlearn’ the old one

A study in the Journal of Cosmology and Astroparticle Physics explores how a machine-learning strategy known as transfer learning could dramatically reduce the computational cost of searching for new physics beyond the standard cosmological model—while also revealing an unexpected risk: Sometimes AI systems can become too reliant on what they already know.

Artificial intelligence is widely used in cosmology to analyze the universe. But testing theories beyond the standard cosmological model, known as ΛCDM, remains extremely computationally demanding.

Although ΛCDM successfully describes many properties of the universe—from its expansion to the distribution of galaxies—physicists know it is probably incomplete. Recent observations hint that phenomena such as massive neutrinos, modified gravity or evolving dark energy could point toward new physics beyond the current model.

/* */