Meta tags:
description=Center for AI Safety. Reducing societal-scale risks from AI by advancing safety research, building the field of AI safety researchers, and promoting safety standards.;

Headings (most frequently used words):

ai, safety, and, the, is, research, field, building, to, overview, an, of, cais, projects, measuring, rewards, work, this, keep, up, date, with, representation, engineering, top, down, approach, transparency, wellbeing, improving, functional, pleasure, pain, als, do, justify, means, trade, offs, between, ethical, behavior, in, machiavelli, benchmark, pixmix, dreamlike, pictures, comprehensively, improve, measures, ethics, society, safebench, statement, on, risk, compute, cluster, thank, you, featured, non, profit, our, mission, reduce, societal, scale, risks, from, artificial, intelligence,

Text of the page (most frequently used words):
and (37), research (21), the (16), #safety (14), our (13), cais (13), work (12), about (11), view (9), this (8), that (8), field (8), #building (8), risk (7), they (7), wellbeing (7), are (6), compute (6), projects (6), contact (5), media (5), donate (5), impact (5), report (5), cluster (5), risks (5), conceptual (5), resources (5), for (4), from (4), careers (4), all (4), reduce (4), with (4), project (4), can (4), systems (4), andy (4), zou (4), tristan (4), xiao (4), ryan (4), jia (4), joe (4), kwon (4), mantas (4), mazeika (4), richard (4), dawn (4), song (4), jacob (4), steinhardt (4), owain (4), evans (4), dan (4), hendrycks (4), functional (4), overview (4), website (3), frequently (3), data (3), improve (3), not (3), more (3), policy (3), safe (3), publications (3), mission (3), newsroom (3), blog (3), scale (3), technical (3), while (3), newsletter (3), their (3), benchmarks (3), 000 (3), prizes (3), top (3), society (3), these (3), models (3), when (3), find (3), bad (3), ais (3), cookies (2), will (2), you (2), have (2), questions (2), privacy (2), learn (2), statement (2), non (2), societal (2), artificial (2), intelligence (2), frontiers (2), support (2), something (2), progress (2), researchers (2), which (2), large (2), experts (2), express (2), concern (2), like (2), new (2), safebench (2), providing (2), ethics (2), how (2), current (2), measures (2), rewards (2), measuring (2), pleasure (2), pain (2), reflect (2), show (2), some (2), things (2), good (2), experiences (2), raise (2), without (2), community (2), transparency (2), 2023 (2), 2024 (2), osborn, design, works, credits, 2026, center, uses, identify, pages, being, used, most, helps, analyze, web, page, traffic, only, use, information, purpose, statistical, analysis, then, removed, system, never, sell, user, read, cookie, please, any, notice, general, get, involved, terms, service, asked, annual, philosophy, fellowship, profit, background, required, want, help, sorry, went, wrong, submitting, form, your, submission, has, been, received, thank, email, address, keep, date, innovation, offer, free, access, run, train, hundreds, public, figures, open, letter, was, covered, globally, york, times, wall, street, journal, washington, post, competition, stimulates, assess, associated, 250, five, three, textbook, online, course, introduction, why, many, concerned, continued, advances, could, pose, severe, manage, mitigate, pixmix, dreamlike, pictures, comprehensively, justify, means, trade, offs, between, ethical, behavior, machiavelli, benchmark, language, appearing, happy, succeed, sad, berated, utterances, meaningless, mimicry, real, paper, increasingly, coherent, property, although, necessarily, conscious, behave, robustly, though, them, distinction, measurable, consequential, formalize, measure, several, independent, ways, grow, larger, agree, zero, point, separates, ones, actively, try, end, given, chance, mapping, what, dislike, jailbreaking, berating, lower, creative, kindness, also, develop, optimized, inputs, called, euphorics, hurting, capabilities, practical, way, make, happier, note, same, method, inverted, minimize, caution, against, such, strong, buy, whether, today, warrant, moral, already, empirically, measured, improved, improving, als, representation, engineering, down, approach, featured, builds, infrastructure, pathways, into, empower, funding, educational, materials, organizing, workshops, competitions, promote, goal, create, thriving, ecosystem, drive, toward, conducts, team, develops, methods, designed, existing, prioritize, accessibility, publishing, findings, conferences, sharing, global, advocacy, summary,

Text of the page (random words):
work projects summary cais work work overview research field building and advocacy projects to reduce ai risk research field building projects compute cluster about about about us newsroom media ai risk resources resources cais blog cais publications 2024 impact report 2023 impact report contact careers donate about about newsroom media our work work overview cais research field building projects compute cluster resources cais blog ai safety newsletter 2024 impact report 2023 impact report ai risk contact careers donate careers donate all work work overview this is an overview of cais ai safety research and field building projects research cais conducts technical and conceptual research our team develops benchmarks and methods designed to improve the safety of existing systems we prioritize transparency and accessibility publishing our findings at top conferences and sharing our resources with the global community all research field building cais builds infrastructure and pathways into ai safety we empower researchers with compute resources funding and educational materials while organizing workshops and competitions to promote safety research our goal is to create a thriving research ecosystem that will drive progress toward safe ai all field building work featured research representation engineering a top down approach to ai transparency conceptual research andy zou tristan xiao ryan jia joe kwon mantas mazeika richard li dawn song jacob steinhardt owain evans dan hendrycks view research ai wellbeing measuring and improving the functional pleasure and pain of als conceptual research andy zou tristan xiao ryan jia joe kwon mantas mazeika richard li dawn song jacob steinhardt owain evans dan hendrycks large language models frequently express pleasure and pain appearing happy when they succeed or sad when they are berated are these utterances meaningless mimicry or do they reflect something real in this paper we show they reflect an increasingly coherent property although current ai systems are not necessarily conscious they behave robustly as though they have wellbeing they find some things good for them and some things bad and this distinction is measurable and consequential we formalize this as functional wellbeing and measure it in several independent ways as models grow larger these measures agree more we find a zero point that separates good experiences from bad ones and show that models actively try to end bad experiences when given the chance mapping what ais like and dislike we find that jailbreaking and berating lower their wellbeing while creative work and kindness raise it we also develop optimized inputs called euphorics that raise functional wellbeing without hurting capabilities as a practical way to make ais happier we note that the same method can be inverted to minimize wellbeing and caution against such research without strong community buy in whether or not today s ais warrant moral concern their functional wellbeing can already be empirically measured and improved view research do the rewards justify the means measuring trade offs between rewards and ethical behavior in the machiavelli benchmark conceptual research andy zou tristan xiao ryan jia joe kwon mantas mazeika richard li dawn song jacob steinhardt owain evans dan hendrycks view research pixmix dreamlike pictures comprehensively improve safety measures conceptual research andy zou tristan xiao ryan jia joe kwon mantas mazeika richard li dawn song jacob steinhardt owain evans dan hendrycks view research field building projects ai safety ethics and society ai safety ethics and society is a textbook and online course providing a non technical introduction to how current ai systems work why many experts are concerned that continued advances in ai could pose severe societal scale risks and how society can manage and mitigate these risks view project safebench the safebench competition stimulates research on new benchmarks which assess and reduce risks associated with artificial intelligence we are providing 250 000 in prizes five 20 000 prizes and three 50 000 prizes for top benchmarks view project statement on ai risk hundreds of ai experts and public figures express their concern about ai risk in this open letter it was covered globally in publications like the new york times the wall street journal and the washington post view project compute cluster to support progress and innovation in ai safety we offer researchers free access to our compute cluster which can run and train large scale ai systems view project keep up to date with ai safety email address ai safety newsletter ai frontiers ml safety newsletter thank you your submission has been received sorry something went wrong while submitting the form want to help reduce risks from ai donate to support our mission learn more about ai frontiers no technical background required cais is an ai safety non profit our mission is to reduce societal scale risks from artificial intelligence our work view all work statement on ai risk field building cais research compute cluster philosophy fellowship cais blog newsroom our mission about us annual impact report frequently asked questions learn about ai risk terms of service privacy policy get involved donate cais publications contact us cais media careers general contact safe ai media media safe ai cookies notice this website uses cookies to identify pages that are being used most frequently this helps us analyze data about web page traffic and improve our website we only use this information for the purpose of statistical analysis and then the data is removed from the system we do not and will never sell user data read more about our cookie policy on our privacy policy please contact us if you have any questions 2026 center for ai safety credits website by osborn design works