Lifeboat Foundation: Safeguarding Humanity
Toggle light / dark theme

Blog

Category: information science

Cloud-tested quantum noise model predicts superconducting qubit errors with sevenfold better accuracy

Researchers from the Johns Hopkins Applied Physics Laboratory (APL) in Laurel, Maryland, and Johns Hopkins University in Baltimore have developed a practical, comprehensive noise-modeling framework for a popular class of superconducting quantum processors. Their work, published in the journal PRX Quantum, offers a sevenfold improvement in predictive accuracy over existing approaches.

Quantum bits, or qubits, are intrinsically prone to noise—interference arising from environmental factors such as electrical and magnetic fields or temperature fluctuations—as a result of the extreme sensitivity that makes them so valuable for computing. Developing accurate noise models is key to creating the robust quantum algorithms and resilient error-correction protocols required to build truly fault-tolerant quantum computers.

“To really advance the field, we need models that can predict a wide range of behavior while utilizing a small number of parameters, rather than theoretical models that try to account for all of the fundamental physics at play in quantum interactions,” said project lead Gregory Quiroz, a senior physicist at APL and an associate research professor in the Department of Physics and Astronomy at the Johns Hopkins University Krieger School of Arts and Sciences. “The novelty of our approach lies in a unified and experimentally validated framework that connects multiple noise mechanisms and yields a coherent predictive methodology.”

Ultra-thin MoS₂ computer packs 1,400 transistors onto one chip

The rapid advancement and diffusion of artificial intelligence (AI) systems, such as the machine learning models underpinning the functioning of ChatGPT, Gemini and similar platforms, have posed new demands on the electronics engineering industry. In fact, these systems are computationally intensive and consume substantial power, particularly when running on existing devices.

Electronics engineers worldwide have thus been trying to develop new hardware systems that can run machine learning algorithms more energy efficiently, without adversely affecting their performance. One promising approach for reducing power consumption entails the use of two-dimensional (2D) semiconductors, ultrathin materials that have already proved promising for the development of smaller electronics.

Researchers at Nanjing University, Suzhou Laboratory and Huawei Technologies Co. Ltd. recently developed and fabricated a fully functional computer based on the 2D semiconductor molybdenum disulfide (MoS₂).

Claude is Self-Evolving?

In this episode, I break down Anthropic’s research on recursive self-improvement—AI systems that can design and train the next generation with less human help—and why the key battleground is “taste” (choosing goals and next steps). I compare this to evolutionary algorithms and newer examples like DeepMind’s AlphaEvolve, Sakana’s Darwin Gödel Machine, and Karpathy’s AutoResearch, then cover METR Task Horizon and how task length has been doubling. I go through Anthropic’s internal results (Claude writing most merged code, speedup experiments, bug fixes, and a study where models sometimes pick better research next steps), plus the main skepticism: bad productivity metrics, internal-only models, and Goodhart’s Law/reward hacking. I end with an open safety problem where Claude agents closed the gap far faster than humans, and what this means for specifying and checking work.

LINKS:
https://www.anthropic.com/institute/r… voice to text App: whryte.com Website: https://engineerprompt.ai/ RAG Beyond Basics Course: https://prompt-s-site.thinkific.com/c… Signup for Newsletter, localgpt: https://tally.so/r/3y9bb0 Let’s Connect: 🦾 Discord: / discord ☕ Buy me a Coffee: https://ko-fi.com/promptengineering |🔴 Patreon: / promptengineering 💼Consulting: https://calendly.com/engineerprompt/c… 📧 Business Contact: engineerprompt@gmail.com Become Member: http://tinyurl.com/y5h28s6h 💻 Pre-configured localGPT VM: https://bit.ly/localGPT (use Code: PromptEngineering for 50% off). Signup for Newsletter, localgpt: https://tally.so/r/3y9bb0 TIMESTAMP: 00:00 Self Improvement Basics 01:30 Evolutionary Loops Today 03:50 Task Horizon Doubling 05:18 Claude Productivity Claims 08:11 Goodhart’s Law 10:30 Agents as Researchers 12:22 What It Means for You.

My voice to text App: whryte.com.
Website: https://engineerprompt.ai/
RAG Beyond Basics Course:
https://prompt-s-site.thinkific.com/c…
Signup for Newsletter, localgpt:

Let’s Connect:

☕ Buy me a Coffee: https://ko-fi.com/promptengineering.
|🔴 Patreon: / promptengineering.
💼Consulting: https://calendly.com/engineerprompt/c…
📧 Business Contact: engineerprompt@gmail.com.
Become Member: http://tinyurl.com/y5h28s6h.

💻 Pre-configured localGPT VM: https://bit.ly/localGPT (use Code: PromptEngineering for 50% off).

Signup for Newsletter, localgpt:

Continue reading “Claude is Self-Evolving?” | >

AI-Discovered Cognitive Models Reveal Novel Insights into Human and Animal Learning

The problem? Human brains (and animal brains, too) are incredibly complex. While these handcrafted models are great starting points, they often oversimplify things and miss the messy, rich reality of actual behavior. On the flip side, using powerful, flexible AI to analyze data can capture that richness, but AI usually gives us a “black box”—it finds patterns but can’t explain *why* or *how* it found them, leaving scientists to do the heavy lifting of figuring out the rules.

Scientific models are widely used across the natural sciences as an interface between scientific theories and empirical data [1]. Such models play a key role, for example, in the study of human and animal learning, where they express algorithmic hypotheses and relate them to psychology and neuroscience data [2, 3]. These models are traditionally handcrafted by expert researchers based on existing theory or new insights. Such handcrafted models, however, are now known to fall short of capturing the full richness of behavior, even in their narrow domains [4– 7]. An alternative data-driven approach has emerged, seeking to discover new insights by fitting and interpreting flexible models [8– 11]. However, these tools require substantial human effort to derive insight from data, and it has been unclear how to discover new ideas from data efficiently. Here, we present DataDIVER, a general approach for automatically discovering computational models from data, and demonstrate that these models surface novel mechanistic insights into human and animal learning. Our approach delivers models that take the form of short computer programs, which are optimized both to fit data well and to be simple. These programs explicitly connect with existing theoretical frameworks and are readily understandable by human scientists. They can also be used to make novel predictions, some of which we show are borne out in re-analysis of existing data. General-purpose tools for surfacing new ideas from data, especially in combination with the large datasets that are increasingly available in many fields, stand to dramatically accelerate scientific discovery.

The authors have declared no competing interest.

Critical Thalamocortical Coordination Dynamics Track Conscious State Transitions

Abstract Despite substantial progress in identifying neural correlates of consciousness, no unified quantitative framework currently derives a formally specified order parameter for conscious-state organisation from established neurophysiological principles, or links thalamocortical coordination dynamics to measurable state transitions across pharmacological, pathological, and perturbational conditions through a single computational formalism. We propose a neurocomputational theoretical framework in which conscious states are associated with metastable regimes of large-scale thalamocortical coordination operating near critical dynamical boundaries. The framework is formalised through a dynamic coordination functional Φ(t), defined as a surface integral over the thalamocortical interface and directly operationalisable from high-density EEG as a weighted combination of gamma-band power spectral density, thalamocortical coherence, and theta-gamma phase-amplitude coupling. The thalamic reticular nucleus (TRN) is identified as the anatomical implementation of the control parameter governing proximity to the critical point, grounded in a Wilson-Cowan model of TRN inhibitory gating whose bifurcation structure is characterised computationally. Numerical simulation of the linearised field equation on the thalamocortical boundary demonstrates internal consistency: the simulated system produces power-law recovery dynamics tau_rec proportional to | θ — θ _c|^v with nu consistent with model A universality class [0.5, 1.5], and a Kuramoto mean-field derivation establishes that Φ(t) emerges as the natural order parameter of coupled thalamocortical oscillators rather than being postulated. The joint (|Φ(t)|, Var[|Φ(t)|]) phase space correctly separates simulated waking, anaesthetic, ictal, and minimally conscious regimes without parameter fitting to empirical data. All simulation code is publicly available. Six quantitatively specific, independently falsifiable predictions are derived across five experimental domains: power-law Gamma Dip scaling in near-threshold EEG with a specific exponent range; causal disruption of thalamocortical coherence by selective TRN silencing; opposite EEG scaling exponent deviations in ASD versus schizophrenia; systematic Φ_est collapse under propofol anaesthesia correlated with PCI; Φ_est as a real-time consciousness biomarker in disorders of consciousness; and clinical validity of Φ_est in disorders of consciousness and ictal state discrimination by the metastability index. Each prediction is stated with quantitative thresholds and a pre-specified falsification criterion. The framework provides: the first anatomically specified and formally derived order parameter for conscious-state organisation directly operationalisable from passive EEG; a mechanistically grounded identification of the TRN as the dynamical control parameter, testable by a single optogenetic experiment; and a computationally validated, pre-registerable programme of six falsifiable predictions defining a tractable empirical agenda. Φ_est would constitute a candidate real-time consciousness biomarker if the framework’s predictions are confirmed in purpose-designed experiments.

Cutting a photon in two creates an infinite swarm of particles

By definition, elementary particles can’t be broken into smaller pieces. But in a new theoretical study published in Physical Review Letters, Johannes Skaar and colleagues have revealed what would happen if you tried anyway for a single photon. The answer is deeply strange: attempting to cut a photon in two wouldn’t produce two smaller photons, but instead conjure an infinite number of them out of thin air.

Like any quantum particle, a photon exists simultaneously as a single, localized particle, and an extended wave, spread out across space. For their investigation, Skaar’s team considered what would happen if a single photon passed through an optical shutter—essentially a very fast mirror that can be switched on and off to block part of a pulse of light. If the shutter was fast enough, it could intercept the photon mid-pulse, snipping off part of this extended wave.

To find out what would happen afterward, the researchers applied quantum equations that describe how the photon’s underlying electromagnetic field behaves at the quantum level. Specifically, their analysis tracked precisely how the photon’s quantum state would be transformed by the shutter’s intervention.

AI and ultralow-energy lasers enable an ultrafast authentication system

The security of modern communications heavily relies on systems that can rapidly and reliably verify users and the devices they are using. This process, known as authentication, essentially entails confirming that users or devices are legitimate (i.e., who or what they claim to be).

Conventional authentication systems rely on static cryptographic keys, fixed digital keys that allow encryption algorithms to scramble readable data into unreadable texts or vice versa. While these systems perform well in some contexts, they often struggle when networks include billions of devices that continuously connect and disconnect.

Researchers at King Abdullah University of Science and Technology (KAUST) recently developed a new system that could authenticate devices faster and more reliably in real time, even when they are connecting to large-scale networks, cloud services or virtual environments.

Ex-OpenAI Scientist WARNS: You Have No Idea What’s Coming In 5 Years

Ilya Sutskever, co-founder of OpenAI and founder of Safe Superintelligence, says the scaling era from 2020 to 2025 is over, that pre-training will run out of data, and that the industry is back to pure research with more companies than ideas. He argues that AGI is the wrong target what is actually coming is a learning algorithm that can take any job, learn it on the fly, and merge that knowledge across millions of simultaneous instances in a way humans cannot, producing rapid economic growth that regulation is unlikely to stop.

He predicts that once AI becomes visibly powerful, frontier companies will become paranoid overnight and governments will scramble, and says the only thing worth building is an AI aligned to sentient life broadly — not human life alone — because the AI itself will be sentient and will vastly outnumber humans within 5 to 20 years.

📚 Sources cited in this video:

Safe Superintelligence Inc. – Company Overview https://ssi.inc.

OpenAI, Founding Charter and Mission https://openai.com/charter.

Ilya Sutskever, Google Scholar – Research Publications https://scholar.google.com/citations?…

⚠️ DISCLAIMER: This channel provides AI commentary and analysis for educational and informational purposes only. Views expressed by guests are their own and do not represent the positions of any company or institution. We encourage viewers to consult multiple sources and form their own conclusions. #ai #agi #artificialintelligence.

Continue reading “Ex-OpenAI Scientist WARNS: You Have No Idea What’s Coming In 5 Years” | >

Paul Dirac

From that insight, Dirac built an entirely new formulation of the theory using what he called “q-numbers” (quantum numbers)—abstract quantities that don’t commute. He independently rediscovered aspects of Hilbert’s operator theory, though he preferred his own algebraic route because he found mathematicians’ obsession with convergence and existence theorems unappealing.

Paul Adrien Maurice Dirac (, dih-RAK ; [ 3 ] 8 August 1902 – 20 October 1984) was a British theoretical physicist who is considered to be one of the founders of quantum mechanics. [ 4 ] [ 5 ] Dirac laid the foundations for both quantum electrodynamics and quantum field theory, coining the former term. [ 6 ] [ 7 ] [ 8 ] [ 9 ] He was Lucasian Professor of Mathematics at the University of Cambridge from 1932 to 1969, and a professor of physics at Florida State University from 1970 to 1984. Dirac shared the 1933 Nobel Prize in Physics with Erwin Schrödinger “for the discovery of new productive forms of atomic theory.” [ 10 ]

Dirac graduated from the University of Bristol with a Bachelor of Science in Electrical Engineering in 1921, and a Bachelor of Arts in Mathematics in 1923. [ 11 ] Dirac then graduated from St John’s College, Cambridge, with a Doctor of Philosophy in Physics in 1926, writing the first ever thesis on quantum mechanics. [ 12 ]

He formulated the Dirac equation, one of the most important results in physics, in 1928. [ 7 ] It connected special relativity and quantum mechanics and predicted the existence of antimatter. [ 13 ] He wrote a famous paper in 1931, [ 14 ] which further predicted the existence of antimatter. [ 15 ] [ 16 ] [ 13 ] Dirac also contributed greatly to the reconciliation of general relativity with quantum mechanics. He contributed to Fermi–Dirac statistics, which describes the behaviour of fermions, particles with half-integer spin. His 1930 monograph, The Principles of Quantum Mechanics, is one of the most influential texts on the subject. [ 17 ] He and Schrödinger tied for eighth in a Physics World poll of the greatest physicists of all time. [ 18 ] .

/* */