The AI & Quantum Computing Chronicle
описание
This channel covers Artificial Intelligence, Data Science, Machine Learning & Quantum Computing to help you extract valuable information through our posts. For any suggestion/question: Twitter: @ItalyHighTech/@KevinClarity Telegram: @vzocca/@kcorella
1 716
подписчиков
Охват к подписчикам
86,1%
ERR
Реакции к просмотрам
0,17%
49 на 20 постов
Пересылки к просмотрам
0,40%
117
Постов в день
0,1
всего 21
Где отзываются чаще
доля реакций к просмотрам- 17 авг.One of the major challenges in quantum computing is not only developing new algorithms, but rigorously demonstrating when they actually provide an advantage over the best classical methods. This work introduces an open benchmarking library with 10 classes of combinatorial optimization problems, including Independent Set, LABS, Network Design, Vehicle Routing, and Portfolio Optimization. The instances are designed to enable systematic comparisons between classical and quantum methods, ranging from problems accessible to current hardware to others that are already difficult for state-of-the-art classical solvers. One of the paper’s most important contributions is its distinction between system benchmarking, algorithm benchmarking, and application benchmarking. Only the latter can support a real demonstration of quantum advantage, because it requires comparing solutions on the same problem without restricting the model, algorithm, or computational platform in advance. The conclusion is particularly cautious: practical demonstrations of quantum advantage in optimization are still missing. The contribution of this work is to create a common scientific infrastructure for measuring progress in a fair, reproducible, and transparent way. In Portfolio Optimization, the benchmark incorporates a multiperiod formulation with risk, return, transaction costs, short selling, and capital constraints, illustrating why dynamic resource allocation remains one of the most demanding problems for both classical and quantum optimization. Link https://www.nature.com/articles/s43588-026-00991-10,43%
- 17 июл.Large language models learn by building associations between words, encoded as relationships in a high-dimensional embedding space. Like Hebb's principle in neuroscience, correlated activity strengthens connections, bringing related concepts closer together. But association alone cannot tell a system when reality has moved beyond its model. Faced with something genuinely novel, it does not recognise the boundary of its knowledge. It interpolates. The question, then, is what makes a world model open rather than closed. Biological intelligence appears to rely on more than association alone. It continually assigns value: this matters, this is dangerous, this is surprising, this deserves to be learnt. Those valuation signals may do more than guide behaviour. They may be what allows a world model to recognise when it no longer fits reality, and to change accordingly. If that's true, then emotion is not the opposite of reason, nor merely an evolutionary add-on. It is part of the architecture that makes intelligence capable of extending itself beyond what it already knows. This essay develops that idea through Hebbian learning, reward prediction error, hallucination, black swan events, and the distinction between solving problems within an existing mathematical framework and creating an entirely new one. Intelligence may not just be developing a world model, but also recognising when the map is no longer adequate. If so, simply scaling a closed world model may never be enough. https://vzocca.substack.com/p/feeling-right-why-emotions-may-be0,42%
- 31 дек.From this channel, we wish you a 2026 full of challenges that sharpen our thinking, projects that demand the best of us, and experiments that reward depth, clarity, and discipline.0,36%
- 1 маяWhat is a decision? It seems trivial until you try to define it. Decision theory explains how we choose, but rarely what a decision is. Choosing is just the outcome. Ronald A. Howard defined it as an allocation of resources. But what truly drives a decision is uncertainty. If the future were known, we would not decide. Warren B. Powell goes further: a decision is a form of information. It defines how we respond to a world we do not fully control. The real challenge is not choosing, but identifying what decisions actually exist. In complex systems, many remain invisible. In AI and quantum systems, this becomes critical. It is no longer about choosing actions, but building structures that decide under uncertainty. Link https://castle.princeton.edu/makingdecisions/0,33%
- 22 мар.This paper provides a complete, ground-up explanation of how Large Language Models (LLMs) work. A central section walks through the entire forward pass with concrete numbers: embedding lookup, Query/Key/Value matrices, the dot-product score matrix, softmax weights, weighted Value sums, layer-by-layer refinement, and the final word prediction. It explains how backpropagation adjusts every matrix in response to prediction errors. It also covers fine-tuning and RLHF in detail. The paper concludes by applying this understanding to explain precisely why LLMs fail at constraint satisfaction. https://vzocca.substack.com/p/how-large-language-models-work0,29%
- 29 дек.How diamonds are powering a new quantum revolution Exactly a century after German scientist Werner Heisenberg built a mathematical framework for explaining quantum physics, the world is launching into what scientists call a “second quantum revolution”. This is where the diamond — the hardest naturally occurring substance on Earth — comes in. It is resistant to vibration because of its rigid crystal lattice of carbon atoms, linked by strong chemical bonds. Most of the carbon atoms have properties intrinsic to their nuclei that make for a magnetically “quiet” environment for quantum effects to take place. https://tinyurl.com/2hdv652e0,24%
- 8 июн.In the last few days we have read about how an AI disproved a conjecture that had stood since 1946 (https://www.scientificamerican.com/article/ai-just-solved-an-80-year-old-erdos-problem-and-mathematicians-are-amazed/). What they don't tell you is that It could not tell you whether its own proof was right. That gap is the whole story, and most of the coverage missed it. Recently an OpenAI model produced a counterexample to the Erdős unit distance conjecture, open for almost 80 years. Nine leading mathematicians verified it. The conjecture is refuted. A real result, and widely misread. Three things that rarely survive the headline: → It could not check its own work. Humans did that. The model had no way to tell this result apart from the confident, plausible-but-wrong proofs these systems produce all the time, and nobody has said how many attempts it took. → It invented no new mathematics. Every technique already existed; the originality was the combination. It overturned Erdős's answer while depending entirely on Erdős's question. → It is a disproof, not a solution. We now know the answer is not what Erdős expected. We still do not know what it is. The mathematicians who verified it call it a milestone. We may need to be more cautious. A genuine result? Yes. Intelligence? No. A very fast machine assembling known pieces and, this once, landing on something true: it may be closer to luck than to insight. https://vzocca.substack.com/p/a-discovery-by-accident0,23%
- 14 дек.The article describes a new mathematical result whose proof draws on ideas originating in string theory, even though the problem it resolves is purely mathematical. The work addresses a long-standing impasse in algebraic geometry: how to understand and classify certain highly complex geometric spaces for which classical methods no longer suffice. The key to the breakthrough does not lie in using physics directly, but in applying the concept of duality, originally developed in string theory, to reformulate the problem in an alternative representation where the underlying structure becomes accessible. In this “mirror” representation, calculations that are intractable in the original framework become manageable, and the equivalence between the two descriptions is then established rigorously. The result is considered brilliant because it opens an entirely new path forward, and baffling because its logic does not emerge from the usual mathematical tradition. The article illustrates how physical intuitions can inspire new formal architectures without introducing physics into the final result. https://www.quantamagazine.org/string-theory-inspires-a-brilliant-baffling-new-math-proof-20251212/0,18%
- 7 июн.The Fourier Transform allows us to take a complex signal and decompose it into elementary frequencies. It turns something apparently “chaotic” into an interpretable structure of components. The Quanta article explains this through very clear examples: the ear separating sounds, heat spreading along a rod, JPEG compression, noise filtering, images as 2D functions, and the connection with quantum mechanics through position and momentum. But the deeper lesson goes beyond signals: Fourier teaches us that sometimes the problem is not in the data itself, but in the domain from which we observe it. In traditional Data Science, this is essential: it transforms temporal, spatial, or sequential data into representations where certain patterns become more separable, compressible, and interpretable. It applies to feature engineering, noise reduction, anomaly detection, images, audio, sensors, and more. In quantum computing and Machine Learning, Fourier appears naturally when we speak about amplitudes, waves, changes of basis, and alternative representations. In quantum mechanics, it connects position and momentum, directly relating to the uncertainty principle. And in portfolio management, the analogy is powerful: a portfolio can also look like a chaotic signal made of initiatives, risks, dependencies, value, debt, operational noise, and external shocks. Perhaps looking only at its “visible state” is not enough. Perhaps we need to decompose it into its dominant frequencies: value cycles, risk patterns, recurring frictions, strategic signals, and noise. Link https://www.quantamagazine.org/what-is-the-fourier-transform-20250903/0,16%
- 22 нояб.For decades we have assumed that quantum mechanics needs imaginary numbers. The famous i, which never appears in any real physical measurement, sits at the heart of Schrödinger’s equation and the structure of the theory. But is it truly essential or simply a useful mathematical choice? In 2021 a special kind of Bell-type experiment seemed to settle the matter. A version of quantum theory built only with real numbers could not reproduce certain observed results. The conclusion looked final: without i, quantum mechanics fails. Yet in 2025 several research teams showed that this is not the whole story. By modifying some basic assumptions that we usually take for granted, such as the standard rule for combining quantum systems, they managed to rebuild the entire theory without imaginary numbers and still obtain the same predictions as the complex version. The interesting part is that even though i disappears from the equations, its effects remain. We still see rotations, phases and interference. The real-number formulation behaves as if an imaginary component were hidden inside another mathematical structure. This raises a deep question. Does nature truly require complex numbers or do we prefer them because they make the theory clearer, more elegant and more intuitive? Even the authors of these new models admit that the standard complex formulation still feels the most natural. And that might be the real lesson: mathematics does not only describe physics. Sometimes it reveals the geometry that makes the physical world possible. https://www.quantamagazine.org/physicists-take-the-imaginary-numbers-out-of-quantum-mechanics-20251107/0,16%
- 2 авг.Modular forms are mathematical functions defined by extremely restrictive symmetries. This rigidity makes it possible to reconstruct their global behavior from a fundamental region and to transform complex problems into far more tractable structures. This is why they have played a decisive role in number theory, in the proof of Fermat’s Last Theorem, and in counting problems through generating functions. Their applications extend to quantum physics and string theory, where they help study states and interactions; graph theory, through the construction of expander graphs; and combinatorial problems, where they allow sequences to be encoded through generating functions and exact formulas to be obtained for their coefficients. From a perspective applied to portfolio management, which is not developed in the article, the most fertile idea would be this: a sufficiently strong structure can reduce global complexity, identify invariants, and make many local decisions governable without analyzing each one in isolation. Link https://www.quantamagazine.org/behold-modular-forms-the-fifth-fundamental-operation-of-math-20230921/0,14%
- 28 февр.In 1935, the Austrian physicist Erwin Schrödinger showed the absurdity of common interpretations of quantum mechanics with his famous cat-based thought experiment. The cat is put into a box with vial of poison, which will be released if a radioactive atom decays. If the box remains isolated from its environment, the atom exists in a superposition of both decayed and not-decayed, and until observed, the cat is an undefined state of both dead and alive. In the real world, objects eventually become too complex or interact too much to maintain a superposition, an idea known as decoherence. But there are also extensions to quantum mechanics, known as collapse theories, that suggest that beyond a certain point, a system will inevitably reduce to a classical state, even in isolation. These theories were picked by 4% of researchers as their favourite interpretation of quantum mechanics in a 2025 Nature survey. https://www.scientificamerican.com/article/quantum-physicists-just-supersized-schroedingers-cat/0,14%