Do neural networks build their representations through smooth, gradual refinement, or via more complex computational processes? We investigate this by extending the logit lens to analyze the policy network of Leela Chess Zero, a superhuman chess engine. Although playing strength and puzzle-solving ability improve consistently across layers, capability progression occurs in distinct computational phases with move preferences undergoing continuous reevaluation--move rankings remain poorly correlated with final outputs until late, and correct puzzle solutions found in middle layers are sometimes overridden. This late-layer reversal is accompanied by concept preference analyses showing final layers prioritize safety over aggression, suggesting a mechanism by which heuristic priors can override tactical solutions.

Hassabis received half of the award alongside a colleague, John Jumper, for the design of AlphaFold: an AI tool that can predict the 3D structure of proteins using only their amino acid sequences--something Hassabis describes as a "50-year grand challenge" in the field of biology. Released freely by Google DeepMind for the world to use five years ago, AlphaFold has revolutionized the work of scientists toiling on research as varied as malaria vaccines, human longevity, and cures for cancer, allowing them to model protein structures in hours rather than years. The Nobel Prizes in 2024 were the first in history to recognize the contributions of AI to the field of science. If Hassabis gets his way, they won't be the last. AlphaFold's impact may have been broad enough to win its creators a Nobel Prize, but in the world of AI, it is seen as almost hopelessly narrow.

In this paper, I apply linguistic methods of analysis to non-linguistic data, metaphorically equating one with the other and seeking analogies. The productivity of this approach has been proven within the field of Super Linguistics. I argue that developed by computational linguists word embeddings (made with the algorithm word2vec) can shed light on the features of chess moves. Recently, computational linguistics has made a great progress in natural language processing (NLP). Within this field, tools have been developed for machine analysis of morphology, syntax and semantics. Computational linguistics became a stand-alone research area with its conferences (weblink) and actual fields. The experts have developed general principles of text analysis, optimal methods of word counting.

This paper presents a new approach that integrates deep learning with computational chess, using both the Mixture of Experts (MoE) method and Monte-Carlo Tree Search (MCTS). Our methodology employs a suite of specialized models, each designed to respond to specific changes in the game's input data. This results in a framework with sparsely activated models, which provides significant computational benefits. Our framework combines the MoE method with MCTS, in order to align it with the strategic phases of chess, thus departing from the conventional ``one-for-all'' model. Instead, we utilize distinct game phase definitions to effectively distribute computational tasks across multiple expert neural networks. Our empirical research shows a substantial improvement in playing strength, surpassing the traditional single-model framework. This validates the efficacy of our integrated approach and highlights the potential of incorporating expert knowledge and strategic principles into neural network design. The fusion of MoE and MCTS offers a promising avenue for advancing machine learning architectures.

This paper presents an empirical exploration of non-transitivity in perfect-information games, specifically focusing on Xiangqi, a traditional Chinese board game comparable in game-tree complexity to chess and shogi. By analyzing over 10,000 records of human Xiangqi play, we highlight the existence of both transitive and non-transitive elements within the game's strategic structure. To address non-transitivity, we introduce the JiangJun algorithm, an innovative combination of Monte-Carlo Tree Search (MCTS) and Policy Space Response Oracles (PSRO) designed to approximate a Nash equilibrium. We evaluate the algorithm empirically using a WeChat mini program and achieve a Master level with a 99.41% win rate against human players. The algorithm's effectiveness in overcoming non-transitivity is confirmed by a plethora of metrics, such as relative population performance and visualization results. Our project site is available at https://sites.google.com/view/jiangjun-site/.

An invading force, 10,000 strong, marches through a storm toward a fortress built into the side of a mountain. From a distance, the combatants look like ants -- menacing and alarmingly well organized. They rattle their spears and snarl through teeth that have never known modern dentistry, and when lightning strikes, it reveals their sheer numbers. Volleys of arrows fly, swords find their way to the weak spots around breast plates. Bodies on both sides hit the ground. This bloody affair is the Battle of Helm's Deep, from The Lord of the Rings: The Two Towers.

AI engines utilizing deep learning neural networks provide excellent tools for analyzing traditional board games. Here we are interested in gaining new insights into the ancient game of Go. For that purpose, we need to define new numerical measures based on the raw output of the engines. In this paper, we develop a numerical tool for automated move-by-move performance evaluation in a context-sensitive manner and for recognizing game features. We measure the urgency of a move by the cost of passing, which is the score value difference between the current configuration of stones and after a hypothetical pass in the same board position. Here we investigate the properties of this measure and describe some applications.

The terms Machine Learning and Deep Learning will be often put in the same basket, but what are they and what is their role? To understand these aspects, the first step is their positioning within the larger umbrella of AI (AGI). To address the limitations, characteristics and differences of these fields, it is necessary to know what an algorithm is since it is the raw material of artificial intelligence, and therefore machine learning and deep learning. An algorithm is a set of instructions that solve a problem. These instructions have to be finite, ordered and logical, in other words, they cannot be an infinite number of instructions.

If you're still emotionally wiped out by Avengers Endgame, The Void and ILMxLAB's latest VR entry might soothe your geeky soul. Avengers Damage Control is more than just a mere virtual reality game, like the upcoming Iron Man title for the PlayStation VR. Instead, it's a prime example of what The Void does best: Building large-scale multi-player VR experiences mapped to physical sets. It's a dream come true for anyone who's ever wanted to fight alongside their favorite Marvel superheroes -- just be prepared to shell out $40 to experience it. Suiting up for Damage Control involves strapping on one of The Void's backpack computers, as well as a huge VR headset.

The last few years have seen an explosion of new paradigms in databases. Previously the relational database management system (RDBMS) as epitomized by the likes of Microsoft's SQLServer or Oracles MySQL had been the de facto route for those looking for a database. I touched on the reasons for this, and looked at some of the newer, or re-discovered, alternatives in one of my earlier pieces; in this article I'm going to dig deeper into one of these, the Graph Database, to explore what they can do, and to show some use cases where they shine. Graph Databases, as the name suggests, organize data in the form of a graph, based on the mathematical principle of graph theory. Fundamentally, we can consider a graph as a collection of nodes and edges.