Lifeboat Foundation: Safeguarding Humanity
Toggle light / dark theme

Blog

Category: robotics/AI – Page 15

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.

Transcending the Brain? AI, Radical Brain Enhancement and the Nature of Consciousness

Human Rights, Ethics, and Artificial Intelligence: Challenges for the next 70 Years of the Universal Declaration.

Susan schneider, university of connecticut, department of philosophy.

Transcending the Brain? AI, Radical Brain Enhancement and the Nature of Consciousness.
The views expressed in this video are those of the speaker(s) at the time of recording and do not necessarily reflect those of the Carr-Ryan Center for Human Rights or Harvard Kennedy School. These perspectives have been presented to encourage debate on important public policy challenges.

Claude Fable 5 and Claude Mythos 5

While Mythos 5 remains largely unconstrained for restricted government and trusted enterprise partners, Fable 5 is wrapped in a sophisticated safety perimeter. If Fable 5 detects a prompt drifting toward high-risk vectors—like cyberwarfare exploits, advanced biology, or chemical synthesis—it doesn’t just give a generic “I can’t answer that” error. Instead, the query seamlessly falls back to Claude Opus 4.8 (Anthropic’s next-most capable model) to handle the response safely.

Today we’re launching Claude Fable 5: a Mythos-class1 model that we’ve made safe for general use.

Fable 5’s capabilities exceed those of any model we’ve ever made generally available. It is state-of-the-art on nearly all tested benchmarks of AI capability, showing exceptional performance in software engineering, knowledge work, vision, scientific research, and many other areas. The longer and more complex the task, the larger Fable 5’s lead over our other models.

Releasing a model this capable comes with risks. Without safeguards, Fable 5’s capabilities in areas like cybersecurity could be misused to cause serious damage. We’ve therefore launched the model with safeguards that mean queries on some topics will instead receive a response from our next-most-capable model, Claude Opus 4.8. To release the model both safely and quickly, we’ve tuned these safeguards conservatively—they’ll sometimes catch harmless requests, though they trigger, on average, in less than 5% of sessions. With more capable models arriving in the coming months, we’re working to improve our safeguards and reduce false positives as quickly as we can.

Asynchronous AI cuts computing energy by orders of magnitude while learning continuously

As artificial intelligence systems grow larger and more powerful, their energy demands are rising dramatically. But recent research from the University of Massachusetts Amherst published in Nature Communications suggests that advanced AI capabilities may be achievable with dramatically lower energy consumption.

A team led by Hava Siegelmann, Provost Professor in the Manning College of Information and Computer Sciences at UMass Amherst, has developed a novel AI that more closely mirrors key aspects of how the human brain operates. Siegelmann and her lab have focused on two complementary goals: enabling AI systems to learn continuously in real time rather than only during a fixed training phase, and dramatically reducing the energy required for intelligent computation.

“Current AI systems are extraordinarily powerful, but they are also extraordinarily energy-hungry,” said Siegelmann. “Our work shows that it is possible to design AI that remains highly capable while operating much more efficiently.”

Hidden geometry explains why kernel methods separate complex data so well

Are two sets of data genuinely different, or is it because of randomness? This question, known as the two-sample testing problem, becomes notoriously difficult in modern datasets, because they are often high-dimensional, complex, and differences between them can take countless subtle forms.

“Simply put, we don’t know what differences to look for, the possibilities are bewildering,” says Professor Victor Panaretos at EPFL’s Institute of Mathematics.

To solve the problem, mathematicians have developed the so-called “kernel methods,” which have emerged as powerful solutions, widely used in fields such as genomics, finance, and artificial intelligence.

/* */