Collective Intelligence

https://www.lesswrong.com/posts/YN7PHizHnxinLsKvy/sociallm-a-language-model-design-for-personalised-apps - идея как докрутить Мамбу с SSM блоками, тренируется на данных диалогов, чатов, и форумов, трекает отдельными SSM блоками не только локальный контекст, но и историю сообщений и чтения пользователя, а также того, что пользователь помнит про конкретного собеседника. Возможные бизнес-приложения: - персональные рекомендации, клиентская поддержка, ИИ-учителя, ментальное здоровье, ИИ-друг а-ля InflectionAI, - приложение для разрешения конфликтов - приложение для улучшения коллективного интеллекта команды, как описано тут: https://ojs.aaai.org/index.php/AAAI/article/view/25755 - интерактивный сторителлинг Возможные научные приложения: - Социальные науки, collective intelligence, AI safety (detection and prevention of collusion and deception)

Система экспериментирования в Spotify 🐍 Ранее я уже неоднократно писала про Spotify (рекомендации, глобальная контрольная группа) Сегодня расскажу о том, как они реализируют эксперименты для работы над персонализацией главной страницы/экрана. ЗАПУСК ЭКСПЕРИМЕНТА: 1. Команда фокусируется на двух областях: скорость запуска экспериментов и их качество. 2. Для тестирования используется платформа Home Config, это внутренняя разработка типа configuration-as-a-service. Платформа позволяет “играть” с сортировкой, контентом, картинками и др. 3. Для запуска теста в прод используется платформа Spotify Experimentation Platform, встроенная во всю экосистем продукта. То есть в Home config создается конфигурация для теста, а дальше он попадает в EP для запуска в прод: параметры, группы, метрики и анализ результатов. 4. Для QA и дебаггинга используется инструмент Home QA. Он симулирует запросы пользователей к главному экрану и проверяет работу запланированного эксперимента. КОЛЛАБОРАЦИЯ: Ожидаемо, все команды хотят запускать тесты на главном экране, поэтому важно иметь прозрачные процессы и обмениваться информацией. Для этого: 1. Spotify использует Experimentation Tracker - хаб для всех экспериментов, который отражает детали каждого теста и помогает приоритизировать их на основе текущих бизнес задач. 2. Experiment Validation Assistant - сервис, который автоматически валидирует все A/B тесты на главной по нескольким факторам и далее отправляет результаты валидации в ветку в Slack с выводом, обратной связью и рекомендациями. 🔗 Полная статья. #spotify

Symmetry-simplicity, broken-symmetry-complexity Author: Krakauer, David C. Source: Interface Focus; Vol.: 13; Article No.: 20220075; Issue: 3; DOI: 10.1098/rsfs.2022.0075; 14 April 2023 SFI Taxonomy: Information Theory, Machine Learning and Statistics Abstract: Complex phenomena are made possible when: (i) fundamental physical symmetries are broken and (ii) from the set of broken symmetries historically selected ground states are applied to performing mechanical work and storing adaptive information. Over the course of several decades Philip Anderson enumerated several key principles that can follow from broken symmetry in complex systems. These include emergence, frustrated random functions, autonomy and generalized rigidity. I describe these as the four Anderson Principles all of which are preconditions for the emergence of evolved function. I summarize these ideas and discuss briefly recent extensions that engage with the related concept of functional symmetry breaking, inclusive of information, computation and causality.

An adaptive bounded-confidence model of opinion dynamics on networks Authors: Kan, Unchitta; Michelle Feng and Mason A. Porter Source: Journal of Complex Networks; Vol.: 11; Issue: 1; Article No.: cnac055; DOI: 10.1093/comnet/cnac055; 30 December 2022 SFI Taxonomy: Social Networks Abstract: Individuals who interact with each other in social networks often exchange ideas and influence each other's opinions. A popular approach to study the spread of opinions on networks is by examining bounded-confidence models (BCMs), in which the nodes of a network have continuous-valued states that encode their opinions and are receptive to other nodes' opinions when they lie within some confidence bound of their own opinion. In this article, we extend the Deffuant-Weisbuch (DW) model, which is a well-known BCM, by examining the spread of opinions that coevolve with network structure. We propose an adaptive variant of the DW model in which the nodes of a network can (1) alter their opinions when they interact with neighbouring nodes and (2) break connections with neighbours based on an opinion tolerance threshold and then form new connections following the principle of homophily. This opinion tolerance threshold determines whether or not the opinions of adjacent nodes are sufficiently different to be viewed as 'discordant'. Using numerical simulations, we find that our adaptive DW model requires a larger confidence bound than a baseline DW model for the nodes of a network to achieve a consensus opinion. In one region of parameter space, we observe 'pseudo-consensus' steady states, in which there exist multiple subclusters of an opinion cluster with opinions that differ from each other by a small amount. In our simulations, we also examine the roles of early-time dynamics and nodes with initially moderate opinions for achieving consensus. Additionally, we explore the effects of coevolution on the convergence time of our BCM.

Economics in nouns and verbs Author: Arthur, W. Brian Source: Journal of Economic Behavior & Organization; Vol.: 205; Pp.: 638-647; DOI: 10.1016/j.jebo.2022.10.036; January 2023 SFI Taxonomy: Models, Tools, and Scientific Visualization (Economics) Abstract: Standard economic theory uses mathematics as its main means of understanding, and this brings clarity of reasoning and logical power. But there is a drawback: algebraic mathematics restricts economic modeling to what can be expressed only in quantitative nouns, and this forces theory to leave out matters to do with process, formation, adjustment, and creation-matters to do with nonequilibrium. For these we need a different means of understanding, one that allows verbs as well as nouns. Algorithmic expression is such a means. It allows verbs-processes-as well as nouns-objects and quantities. It allows fuller description in economics, and can include heterogeneity of agents, actions as well as objects, and realistic models of behavior in ill-defined situations. The world that algorithms reveal is action-based as well as object-based, organic, possibly ever-changing, and not fully knowable. But it is strangely and wonderfully alive.

Does big data serve policy? Not without context. An experiment with in silico social science Authors: Graziul, Chris; Belikov, Alexander; Ishanu Chattopadyay; Ziwen Chen; Hongbo Fang; Anuraag Girdhar; Xiaoshuang Jua; P. M. Krafft; Max Kleiman-Weiner; Candice Lewis; Chen Liang; John Muchovej; Alejandro Vietos; Meg Young and James Evans Source: Computational and Mathematical Organizational Theory; Vol.: 29; Issue: 1; Pp.: 188-219; DOI: 10.1007/s10588-022-09362-3; March 2023 SFI Taxonomy: Models, Tools, and Scientific Visualization (Human Social Dynamics) Abstract: The DARPA Ground Truth project sought to evaluate social science by constructing four varied simulated social worlds with hidden causality and unleashed teams of scientists to collect data, discover their causal structure, predict their future, and prescribe policies to create desired outcomes. This large-scale, long-term experiment of in silico social science, about which the ground truth of simulated worlds was known, but not by us, reveals the limits of contemporary quantitative social science methodology. First, problem solving without a shared ontology-in which many world characteristics remain existentially uncertain-poses strong limits to quantitative analysis even when scientists share a common task, and suggests how they could become insurmountable without it. Second, data labels biased the associations our analysts made and assumptions they employed, often away from the simulated causal processes those labels signified, suggesting limits on the degree to which analytic concepts developed in one domain may port to others. Third, the current standard for computational social science publication is a demonstration of novel causes, but this limits the relevance of models to solve problems and propose policies that benefit from the simpler and less surprising answers associated with most important causes, or the combination of all causes. Fourth, most singular quantitative methods applied on their own did not help to solve most analytical challenges, and we explored a range of established and emerging methods, including probabilistic programming, deep neural networks, systems of predictive probabilistic finite state machines, and more to achieve plausible solutions. However, despite these limitations common to the current practice of computational social science, we find on the positive side that even imperfect knowledge can be sufficient to identify robust prediction if a more pluralistic approach is applied. Applying competing approaches by distinct subteams, including at one point the vast TopCoder.comglobal community of problem solvers, enabled discovery of many aspects of the relevant structure underlying worlds that singular methods could not. Together, these lessons suggest how different a policy-oriented computational social science would be than the computational social science we have inherited. Computational social science that serves policy would need to endure more failure, sustain more diversity, maintain more uncertainty, and allow for more complexity than current institutions support.

Автогенерация интерфейсов Новая нейросеть от гугл - Gemini - умеет генерировать интерфейсы внутри чата в зависимости от задачи пользователя, выглядит очень перспективно - как примерно и описывалось в статьях об автогенерации интерфейсов в LLM. Интересно сколько там все-таки заготовленных канвас? https://www.theverge.com/2023/12/6/23990466/google-gemini-llm-ai-model Узнал от @cryptoEssay