Recent advancements in multi-agent reinforcement learning (MARL) have demonstrated its application potential in modern games. Beginning with foundational work and progressing to landmark achievements such as AlphaStar in StarCraft II and OpenAI Five in Dota 2, MARL has proven capable of achieving superhuman performance across diverse game environments through techniques like self-play, supervised learning, and deep reinforcement learning. With its growing impact, a comprehensive review has become increasingly important in this field. This paper aims to provide a thorough examination of MARL's application from turn-based two-agent games to real-time multi-agent video games including popular genres such as Sports games, First-Person Shooter (FPS) games, Real-Time Strategy (RTS) games and Multiplayer Online Battle Arena (MOBA) games. We further analyze critical challenges posed by MARL in video games, including nonstationary, partial observability, sparse rewards, team coordination, and scalability, and highlight successful implementations in games like Rocket League, Minecraft, Quake III Arena, StarCraft II, Dota 2, Honor of Kings, etc. This paper offers insights into MARL in video game AI systems, proposes a novel method to estimate game complexity, and suggests future research directions to advance MARL and its applications in game development, inspiring further innovation in this rapidly evolving field.

Real-time toxicity detection in online environments poses a significant challenge, due to the increasing prevalence of social media and gaming platforms. We introduce ToxBuster, a simple and scalable model that reliably detects toxic content in real-time for a line of chat by including chat history and metadata. ToxBuster consistently outperforms conventional toxicity models across popular multiplayer games, including Rainbow Six Siege, For Honor, and DOTA 2. We conduct an ablation study to assess the importance of each model component and explore ToxBuster's transferability across the datasets. Furthermore, we showcase ToxBuster's efficacy in post-game moderation, successfully flagging 82.1% of chat-reported players at a precision level of 90.0%. Additionally, we show how an additional 6% of unreported toxic players can be proactively moderated.

Esport games comprise a sizeable fraction of the global games market, and is the fastest growing segment in games. This has given rise to the domain of esports analytics, which uses telemetry data from games to inform players, coaches, broadcasters and other stakeholders. Compared to traditional sports, esport titles change rapidly, in terms of mechanics as well as rules. Due to these frequent changes to the parameters of the game, esport analytics models can have a short life-spam, a problem which is largely ignored within the literature. As a case study, a neural network model is trained to predict the number of kills in a Dota 2 match utilising this novel character representation technique. The performance of this model is then evaluated against two distinct baselines, including conventional techniques. Not only did the model significantly outperform the baselines in terms of accuracy (85% AUC), but the model also maintains the accuracy in two newer iterations of the game that introduced one new character and a brand new character type. These changes introduced to the design of the game would typically break conventional techniques that are commonly used within the literature. Therefore, the proposed methodology for representing characters can increase the life-spam of machine learning models as well as contribute to a higher performance when compared to traditional techniques typically employed within the literature. Beyond the Meta: Leveraging Game Design Parameters for Patch-Agnostic Esport Analitics Introduction Esport titles, such as League of Legends and Dota 2, have amassed both large audiences and player-bases (Newzoo, 2022; Petrovskaya and Zendle, 2020). Due to the competitive nature of the genre, the player community often develop so called "metas" as explained by Kokkinakis et al. (2021). According to the author, metas are naturally discovered and developed strategies for optimum ways of playing the game that are focused in determining competitive advantage available within the current parameters of the game design.

OpenAI has released its latest version of the language model, GPT-4, which it calls a "milestone in our effort in scaling up deep learning". While the company credits the achievement to a team effort, for OpenAI's founder Sam Altman, one person stands out as a driving force behind the pretraining effort – Jakub Pachocki. GPT-4 was truly a team effort from our entire company, but the overall leadership and technical vision of Jakub Pachocki for the pretraining effort was remarkable and we wouldn't be here without it Pachocki has been with OpenAI since 2017, and his technical vision and leadership played a crucial role in the development of GPT-4. According to Altman, "we wouldn't be here without him". In a recent interview with MIT, he said "That fundamental formula has not changed much for years," talking about the evolution of GPT models since the first version released in 2018.

Recent work has shown that, in generative modeling, cross-entropy loss improves smoothly with model size and training compute, following a power law plus constant scaling law. One challenge in extending these results to reinforcement learning is that the main performance objective of interest, mean episode return, need not vary smoothly. To overcome this, we introduce *intrinsic performance*, a monotonic function of the return defined as the minimum compute required to achieve the given return across a family of models of different sizes. We find that, across a range of environments, intrinsic performance scales as a power law in model size and environment interactions. Consequently, as in generative modeling, the optimal model size scales as a power law in the training compute budget. Furthermore, we study how this relationship varies with the environment and with other properties of the training setup. In particular, using a toy MNIST-based environment, we show that varying the "horizon length" of the task mostly changes the coefficient but not the exponent of this relationship.

Gone are the days when free games equated to ropey puzzle games and knock-off clones of games that had a price tag. These days, courtesy of in-app payments and the ease of offering both a sample of a title alongside its full-fat version, there are actually plenty of great games to play without spending a dime. Following the continued success of titles like Fortnite, the level of quality across mobile, PC and consoles has never been higher. Put your card away, and consider this your starting point. At launch, many of us assumed Genshin Impact was a tenuous Zelda: Breath of the Wild copy with impressive anime graphics and not much else.

OpenAI is a non-profit research organization dedicated to developing and applying artificial intelligence (AI) for the benefit of humanity as a whole. Elon Musk and Sam Altman founded the company in 2015, headquartered in San Francisco, California. OpenAI was founded partly due to its founders' existential fears about the potential for a disaster caused by carelessness and misuse of general-purpose AI. The company focuses on fundamental advances in artificial intelligence and its capabilities. The company's two founders and other investors began with a $1 billion endowment. Elon Musk left the company in February 2018 due to potential conflicts with his work at Tesla, Nikola Tesla's electronics company.

Video games are not just for fun anymore, thanks to esports (electronic sports), also known as live-streamed professional gaming. In a typical esports competition, teams of expert game players face off against each other in a range of popular titles, like League of Legends and Dota 2. Their every move is watched, scrutinized, and analyzed by millions of viewers digitally logging into live streams, attending live events, or watching match recaps. The top players in the world, often known better by their on-screen handles than their real names, get paid a fortune in prize money. Esports teams play live events in the actual arenas used for traditional sports and rock concerts, like the Crypto.com If this all sounds a little serious for playing games, that is because it is.

The artificial neural network is currently the best technique for solving image detection, sound, and natural language processing problems when a big amount of data is available. Image detection is perhaps the most interesting of those specified before, using convolutional neural networks to learn patterns from data. This is a fairly recent technique with applications in many fields, the most famous being automatic cars, and face detection. Artificial neural networks represent a computational system inspired by nature, more precisely by the functioning of biological neurons in the human brain. The fundamental idea behind neural networks is that if they work in nature, they should also work inside a computer.

Multiplayer Online Battle Arena (MOBA) is one of the most successful game genres. MOBA games such as League of Legends have competitive environments where players race for their rank. In most MOBA games, a player's rank is determined by the match result (win or lose). It seems natural because of the nature of team play, but in some sense, it is unfair because the players who put a lot of effort lose their rank just in case of loss and some players even get free-ride on teammates' efforts in case of a win. To reduce the side-effects of the team-based ranking system and evaluate a player's performance impartially, we propose a novel embedding model that converts a player's actions into quantitative scores based on the actions' respective contribution to the team's victory. Our model is built using a sequence-based deep learning model with a novel loss function working on the team match. The sequence-based deep learning model process the action sequence from the game start to the end of a player in a team play using a GRU unit that takes a hidden state from the previous step and the current input selectively. The loss function is designed to help the action score to reflect the final score and the success of the team. We showed that our model can evaluate a player's individual performance fairly and analyze the contributions of the player's respective actions.