A tertiary source is an index or textual consolidation of already published primary and secondary sources that does not provide additional interpretations or analysis of the sources. Some tertiary sources can be used as an aid to find key (seminal) sources, key terms, general common knowledge and established mainstream science on a topic. The exact definition of tertiary varies by academic field. Academic research standards generally do not accept tertiary sources such as encyclopedias as citations, although survey articles are frequently cited rather than the original publication. == Overlap with secondary sources == As is also the case with distinguishing primary and secondary sources in some disciplines, there is not always a clear distinguishing line between secondary and tertiary sources. Depending on the topic of research, a scholar may use a bibliography, dictionary, or encyclopedia as either a tertiary or a secondary source. This causes some difficulty in defining many sources as either one type or the other. In some academic disciplines, the differentiation between a secondary and tertiary source is relative. In the United Nations International Scientific Information System (UNISIST) model, a secondary source is a bibliography, whereas a tertiary source is a synthesis of primary sources. == Types of tertiary sources == Tertiary sources can come in book form or as an online resource. Tertiary sources in book form are frequently organised in alphabetical order, whereas an online tertiary source may be searchable by keyword. Examples of tertiary sources include; reference books, encyclopedias, dictionaries, some textbooks, abstracts, directories, factbooks, handbooks, manuals and compendia. Indexes, bibliographies, concordances, and databases are aggregates of primary and secondary sources and therefore often considered tertiary sources. They may also serve as a point of access to the full or partial text of primary and secondary sources. Almanacs, travel guides, field guides, and timelines are also examples of tertiary sources. Tertiary sources attempt to summarize, collect, and consolidate the source materials into an overview without adding analysis and synthesis of new conclusions. Wikipedia is a tertiary source.
Symbolic artificial intelligence
In artificial intelligence, symbolic artificial intelligence (also known as classical artificial intelligence or logic-based artificial intelligence) is the term for the collection of all methods in artificial intelligence research that are based on high-level symbolic (human-readable) representations of problems, logic, and search. Symbolic AI used tools such as logic programming, production rules, semantic nets and frames, and it developed applications such as knowledge-based systems (in particular, expert systems), symbolic mathematics, automated theorem provers, ontologies, the semantic web, and automated planning and scheduling systems. The Symbolic AI paradigm led to important ideas in search, symbolic programming languages, agents, multi-agent systems, the semantic web, and the strengths and limitations of formal knowledge and reasoning systems. Symbolic AI was the dominant paradigm of AI research from the mid-1950s until the mid-1990s. Researchers in the 1960s and the 1970s were convinced that symbolic approaches would eventually succeed in creating a machine with artificial general intelligence and considered this the ultimate goal of their field. An early boom, with early successes such as the Logic Theorist and Samuel's Checkers Playing Program, led to unrealistic expectations and promises and was followed by the first AI Winter as funding dried up. A second boom (1969–1986) occurred with the rise of expert systems, their promise of capturing corporate expertise, and an enthusiastic corporate embrace. That boom, and some early successes, e.g., with XCON at DEC, was followed again by later disappointment. Problems with difficulties in knowledge acquisition, maintaining large knowledge bases, and brittleness in handling out-of-domain problems arose. Another, second, AI Winter (1988–2011) followed. Subsequently, AI researchers focused on addressing underlying problems in handling uncertainty and in knowledge acquisition. Uncertainty was addressed with formal methods such as hidden Markov models, Bayesian reasoning, and statistical relational learning. Symbolic machine learning addressed the knowledge acquisition problem with contributions including Version Space, Valiant's PAC learning, Quinlan's ID3 decision-tree learning, case-based learning, and inductive logic programming to learn relations. Neural networks, a subsymbolic approach, had been pursued from early days and reemerged strongly in 2012. Early examples are Rosenblatt's perceptron learning work, the backpropagation work of Rumelhart, Hinton and Williams, and work in convolutional neural networks by LeCun et al. in 1989. However, neural networks were not viewed as successful until about 2012: "Until Big Data became commonplace, the general consensus in the Al community was that the so-called neural-network approach was hopeless. Systems just didn't work that well, compared to other methods. ... A revolution came in 2012, when a number of people, including a team of researchers working with Hinton, worked out a way to use the power of GPUs to enormously increase the power of neural networks." Over the next several years, deep learning had spectacular success in handling vision, speech recognition, speech synthesis, image generation, and machine translation, though symbolic approaches continue to be useful in a few domains such as computer algebra systems and proof assistants. == History == A short history of symbolic AI to the present day follows below. Time periods and titles are drawn from Henry Kautz's 2020 AAAI Robert S. Engelmore Memorial Lecture and the longer Wikipedia article on the History of AI, with dates and titles differing slightly for increased clarity. === The first AI summer: irrational exuberance, 1948–1966 === Success at early attempts in AI occurred in three main areas: artificial neural networks, knowledge representation, and heuristic search, contributing to high expectations. This section summarizes Kautz's reprise of early AI history. ==== Approaches inspired by human or animal cognition or behavior ==== Cybernetic approaches attempted to replicate the feedback loops between animals and their environments. A robotic turtle, with sensors, motors for driving and steering, and seven vacuum tubes for control, based on a preprogrammed neural net, was built as early as 1948. This work can be seen as an early precursor to later work in neural networks, reinforcement learning, and situated robotics. An important early symbolic AI program was the Logic theorist, written by Allen Newell, Herbert Simon and Cliff Shaw in 1955–56, as it was able to prove 38 elementary theorems from Whitehead and Russell's Principia Mathematica. Newell, Simon, and Shaw later generalized this work to create a domain-independent problem solver, GPS (General Problem Solver). GPS solved problems represented with formal operators via state-space search using means-ends analysis. During the 1960s, symbolic approaches achieved great success at simulating intelligent behavior in structured environments such as game-playing, symbolic mathematics, and theorem-proving. AI research was concentrated in four institutions in the 1960s: Carnegie Mellon University, Stanford, MIT and (later) University of Edinburgh. Each one developed its own style of research. Earlier approaches based on cybernetics or artificial neural networks were abandoned or pushed into the background. Herbert Simon and Allen Newell studied human problem-solving skills and attempted to formalize them, and their work laid the foundations of the field of artificial intelligence, as well as cognitive science, operations research and management science. Their research team used the results of psychological experiments to develop programs that simulated the techniques that people used to solve problems. This tradition, centered at Carnegie Mellon University would eventually culminate in the development of the Soar architecture in the middle 1980s. ==== Heuristic search ==== In addition to the highly specialized domain-specific kinds of knowledge that we will see later used in expert systems, early symbolic AI researchers discovered another more general application of knowledge. These were called heuristics, rules of thumb that guide a search in promising directions: "How can non-enumerative search be practical when the underlying problem is exponentially hard? The approach advocated by Simon and Newell is to employ heuristics: fast algorithms that may fail on some inputs or output suboptimal solutions." Another important advance was to find a way to apply these heuristics that guarantees a solution will be found, if there is one, not withstanding the occasional fallibility of heuristics: "The A algorithm provided a general frame for complete and optimal heuristically guided search. A is used as a subroutine within practically every AI algorithm today but is still no magic bullet; its guarantee of completeness is bought at the cost of worst-case exponential time. ==== Early work on knowledge representation and reasoning ==== Early work covered both applications of formal reasoning emphasizing first-order logic, along with attempts to handle common-sense reasoning in a less formal manner. ===== Modeling formal reasoning with logic: the "neats" ===== Unlike Simon and Newell, John McCarthy felt that machines did not need to simulate the exact mechanisms of human thought, but could instead try to find the essence of abstract reasoning and problem-solving with logic, regardless of whether people used the same algorithms. His laboratory at Stanford (SAIL) focused on using formal logic to solve a wide variety of problems, including knowledge representation, planning and learning. Logic was also the focus of the work at the University of Edinburgh and elsewhere in Europe which led to the development of the programming language Prolog and the science of logic programming. ===== Modeling implicit common-sense knowledge with frames and scripts: the "scruffies" ===== Researchers at MIT (such as Marvin Minsky and Seymour Papert) found that solving difficult problems in vision and natural language processing required ad hoc solutions—they argued that no simple and general principle (like logic) would capture all the aspects of intelligent behavior. Roger Schank described their "anti-logic" approaches as "scruffy" (as opposed to the "neat" paradigms at CMU and Stanford). Commonsense knowledge bases (such as Doug Lenat's Cyc) are an example of "scruffy" AI, since they must be built by hand, one complicated concept at a time. === The first AI winter: crushed dreams, 1967–1977 === The first AI winter was a shock: During the first AI summer, many people thought that machine intelligence could be achieved in just a few years. The Defense Advance Research Projects Agency (DARPA) launched programs to support AI research to use AI to solve problems of national security; in particular, to automate the translation of Russian to English for inte
Commercial skipping
Commercial skipping is a feature of some digital video recorders that makes it possible to automatically skip commercials in recorded programs. This feature created controversy, with major television networks and movie studios claiming it violates copyright and should be banned. == History == After the video cassette recorder (VCR) became popular in the 1980s, the television industry began studying the impact of users fast forwarding through commercials. Advertising agencies fought the trend by making them more entertaining. For many years, video recorders manufactured for the Japanese market have been able to skip advertisements automatically, which is done by detecting when foreign language audio overdub tracks provided for many programmes go silent, as advertisements were broadcast with a single language only. The first digital video recorder (DVR) with a built-in commercial skipping feature was ReplayTV with its "4000 Series" and "5000 Series" units. In 2002, the main television networks and movie studios sued ReplayTV, claiming that skipping advertisements during replay violates copyright. Later, five owners of ReplayTV represented by Electronic Frontier Foundation and attorneys Ira Rothken and Richard Wiebe countersued, asking the federal judge to uphold consumers' rights to record TV shows and skip commercials, claiming that features like commercial skipping help parents protect their kids from excessive consumerism. ReplayTV ended up filing for bankruptcy in 2003 after fighting a copyright infringement suit over the ReplayTV's ability to skip commercials. === Commercial skipping software === In addition to the DVR devices which existed in the private market since the late 1990s, towards the mid-2000s, due to the significant advances in home computers, Home theater PCs started gaining popularity in the private market and many users began using their Home theater PCs in their living room for entertainment purposes. Following this, many DVR programs were developed, including popular programs such as Windows Media Center, which contained all of the features of the DVR devices in addition to advanced features such as HDTV and the use of Multiple TV Tuner Cards. Some independent developers began developing independent software capable of skipping the commercial segments when playing recorded videos, and permanently removing the commercial segments from recorded video files. By 2014, many DVR programs such as Windows Media Center, SageTV and MythTV had the capability to skip commercials segments in recorded TV broadcasts after installing third-party add-ons such as DVRMSToolbox, Comskip and ShowAnalyzer, which use various advanced techniques to locate the commercial segments in the video files and save their locations to text files. The text files can also be fed into programs such as MEncoder or DVRMSToolboxGUI which can delete the commercial segments from the recorded video files. A few third-party tools such as MCEBuddy automate detection and removal/marking of commercials. One of the weaknesses of commercial skippers is that, operating automatically, they may misidentify program material as a commercial. Some programs like MCEBuddy provide the ability to fine-tune commercial detection for groups of files (e.g. by channel or country) and provide tools to manually fine-tune commercial segments for individual files. In May 2012, the US Dish Network began offering a DVR with what it calls AutoHop. The device would automatically skip commercials when displaying programming that the viewer had previously recorded with the PrimeTime Anytime feature. It does not skip ads on any live programs. US broadcasters were angered at the news, and FOX embarked on legal action. Most, but not all, of Fox's claims were dismissed; ultimately an agreement was reached whereby AutoHop would only become available for Fox stations seven days after a program is transmitted; terms of the settlement were not disclosed. == The future of TV advertisements == The introduction of digital video recorders and services with skipping and fast-forward capabilities enables viewers to avoid viewing interruptive advertisements in recorded programs, either manually or automatically. While advertising separate to television shows can be skipped, advertising in TV shows themselves ("product placement") cannot be skipped. Streaming services such as Hulu show shorter advertisements with a countdown timer and tailored to the viewers interests, asking interactive questions like "Is this ad relevant to you?".
Digital divide
Digital divide is inequitable access to and use of digital technology, encompassing four interrelated dimensions: motivational, material, skills, and usage access. The digital divide worsens inequality in access to information and resources. According to 2026 data from the U.S. Census Bureau, a significant 'digital divide' persists, with over 15.7 million Americans lacking access to high-speed broadband. Students from low-income households often face limited access to reliable internet and digital devices, which negatively affects their educational opportunities. In the Information Age, people without access to the Internet and other technology are at a disadvantage, for they are less able to connect with others, find and apply for jobs, shop, and learn. People living in poverty, in insecure housing or who are homeless, elderly people, and those living in rural communities may have limited access to the Internet; in contrast, urban middle class people have easy access to the Internet. Another divide is between producers and consumers of Internet content, which could be a result of educational disparities. While social media use varies across age groups, a US 2010 study reported no racial divide. == History == The historical roots of the digital divide in the United States refer to the increasing gap that occurred during the early modern period between those who could and could not access the real time forms of calculation, decision-making, and visualization offered via written and printed media. "Over time, focus has shifted from binary access to differentiated use, where quality and purpose of engagement vary across socio-economic groups." Within this context, ethical discussions regarding the relationship between education and the free distribution of information were raised by thinkers such as Immanuel Kant, Jean Jacques Rousseau, and Mary Wollstonecraft (1712–1778). The latter advocated that governments should intervene to ensure that any society's economic benefits should be fairly and meaningfully distributed. Amid the Industrial Revolution in Great Britain, Rousseau's idea helped to justify poor laws that created a safety net for those who were harmed by new forms of production. Later, when telegraph and postal systems evolved, many used Rousseau's ideas to argue for full access to those services, even if it meant subsidizing hard-to-serve citizens. Thus, "universal services" referred to innovations in regulation and taxation that would allow phone services such as AT&T in the United States to serve hard-to-serve rural users. In 1996, as telecommunications companies merged with Internet companies, the Federal Communications Commission adopted Telecommunications Act of 1996 to consider regulatory strategies and taxation policies to close the digital divide. Though the term "digital divide" was coined among consumer groups that sought to tax and regulate information and communications technology (ICeT) companies to close the digital divide, the topic soon moved onto a global stage. The focus was the World Trade Organization which passed the Telecommunications Services Act, which resisted regulation of ICT companies so that they would be required to serve hard-to-serve individuals and communities. In 1999, to assuage anti-globalization forces, the WTO hosted the "Financial Solutions to Digital Divide" in Seattle, US, co-organized by Craig Warren Smith of Digital Divide Institute and Bill Gates Sr. the chairman of the Bill and Melinda Gates Foundation. It catalyzed a full-scale global movement to close the digital divide, which quickly spread to all sectors of the global economy. In 2000, US president Bill Clinton mentioned the term in the State of the Union Address. Since the early 2000s, the international community has transitioned from a focus on domestic infrastructure to a global, multi-dimensional framework for digital equity. This shift was formalized through the World Summit on the Information Society (WSIS) in Geneva (2003) and Tunis (2005), where the International Telecommunication Union (ITU) established a roadmap for bridging the Global North-South disparity as part of the Sustainable Development Goals. Academic and policy discourse has since evolved to distinguish between the first-level divide (physical access), the second-level divide (digital literacy), and the third-level divide (the ability to translate technology use into socio-economic capital). By the 2020s, critical reflections on national development emphasized that the divide is fundamentally a socio-institutional gap. Research by Tiwari, Kostenko, and Yekhanurov (2025) identifies four pillars for achieving national digital maturity which are digital governance capacity, institutional design to prevent adverse digital incorporation, infrastructure resilience, and citizen capability. This modern era is characterized by the pursuit of meaningful connectivity, a standard that requires internet access to be not only available but affordable, high-speed, and supportive of active content creation. === During the COVID-19 pandemic === At the outset of the COVID-19 pandemic, governments worldwide issued stay-at-home orders that imposed lockdowns, quarantines, restrictions, and closures. The resulting interruptions to schooling, public services, and business operations drove nearly half of the world's population into seeking alternative methods to live while in isolation. These methods included telemedicine, virtual classrooms, online shopping, technology-based social interactions and working remotely, all of which require access to high-speed or broadband internet access and digital technologies. A Pew Research Centre study reports that 90% of Americans describe the use of the Internet as "essential" during the pandemic. The accelerated use of digital technologies created a landscape where the ability, or lack thereof, to access digital spaces became a crucial factor in everyday life. According to the Pew Research Center, 59% of children from lower-income families were likely to face digital obstacles in completing school assignments. These obstacles included the use of a cellphone to complete homework, having to use public Wi-Fi because of unreliable internet service in the home and lack of access to a computer in the home. This difficulty, titled the homework gap, affects more than 30% of K-12 students living below the poverty threshold, and disproportionally affects American Indian/Alaska Native, Black, and Hispanic students. These types of interruptions or privilege gaps in education exemplify problems in the systemic marginalization of historically oppressed individuals in primary education. The pandemic exposed inequity causing discrepancies in learning. "Large-scale events such as COVID-19 intensify both access and skills gaps, underlining the need for resilient digital inclusion policies. Studies during COVID-19 reveal first-level (access) and second-level (skills) divides, with underserved students struggling with reliable internet, devices, and platform navigation ” A lack of "tech readiness", that is, confident and independent use of devices, was reported among the US elderly population; with more than 50% reporting an inadequate knowledge of devices and more than one-third reporting a lack of confidence. "Older adults often face skills and confidence barriers, illustrating later-stage divides in van Dijk’s model." Moreover, according to a UN research paper, similar results can be found across various Asian countries, with those aged over 74, reporting less confident or inconsistent use of digital devices. This aspect of the digital divide and the elderly occurred during the pandemic as healthcare providers increasingly relied upon telemedicine to manage chronic and acute health conditions. == Aspects == There are various definitions of the digital divide, all with slightly different emphasis, which is evidenced by related concepts like digital inclusion, digital participation, digital skills, media literacy, and digital accessibility.“Van Dijk’s model identifies sequential barriers—motivational, material, skills, and usage—that must be addressed to bridge the divide.” === Infrastructure === The infrastructure by which individuals, households, businesses, and communities connect to the Internet addresses the physical mediums that people use to connect to the Internet such as desktop computers, laptops, basic mobile phones or smartphones, MP3 players, gaming consoles, electronic book readers, and tablets. Traditionally, the nature of the divide has been measured in terms of the existing numbers of subscriptions and digital devices. Given the increasing number of such devices, some have concluded that the digital divide among individuals has increasingly been closing as the result of a natural and almost automatic process. Others point to persistent lower levels of connectivity among women, racial and ethnic minorities, people with lower incomes, rura
Audience capture
Audience capture is the phenomenon where an influencer is affected by their audience, catering to it with what they believe it wants to hear or is willing to pay for. This creates a positive feedback loop, which can lead the influencer to express more extreme views and behaviors. A famous example of audience capture can be found in the story of the online influencer Nicholas Perry, known as Nikocado Avocado. Perry started off on YouTube with videos of himself playing the violin and supporting veganism. He then shifted to videos of himself eating known as mukbang. Audience capture led him to more and more extreme eating leading him in turn to obesity and poor health. The effect can cause ideological media creators to become more politically radical, based on the feedback of their audience.
AI safety
AI safety is an interdisciplinary field focused on preventing accidents, misuse, or other harmful consequences arising from artificial intelligence systems. It encompasses AI alignment (which aims to ensure AI systems behave as intended), monitoring AI systems for risks, and enhancing their robustness. The field is particularly concerned with existential risks posed by advanced AI models. Beyond technical research, AI safety involves developing norms and policies that promote safety, including advocacy for regulations at different levels of government. The field gained significant popularity in 2023, with rapid progress in generative AI and public concerns voiced by researchers and CEOs about potential dangers. During the 2023 AI Safety Summit, the United States and the United Kingdom both established their own AI Safety Institute. However, researchers have expressed concern that AI safety measures are not keeping pace with the rapid development of AI capabilities. == Motivations == Scholars discuss current risks from critical systems failures, bias, and AI-enabled surveillance, as well as emerging risks like technological unemployment, digital manipulation, weaponization, AI-enabled cyberattacks and bioterrorism. They also discuss speculative risks from losing control of future artificial general intelligence (AGI) agents, or from AI enabling perpetually stable dictatorships. === Existential safety === Some have criticized concerns about AGI, such as Andrew Ng who compared them in 2015 to "worrying about overpopulation on Mars when we have not even set foot on the planet yet". Stuart J. Russell on the other side urges caution, arguing that "it is better to anticipate human ingenuity than to underestimate it". AI researchers have widely differing opinions about the severity and primary sources of risk posed by AI technology – though surveys suggest that experts take high consequence risks seriously. In two surveys of AI researchers, the median respondent was optimistic about AI overall, but placed a 5% probability on an "extremely bad (e.g. human extinction)" outcome of advanced AI. In a 2022 survey of the natural language processing community, 37% agreed or weakly agreed that it is plausible that AI decisions could lead to a catastrophe that is "at least as bad as an all-out nuclear war". == History == Risks from AI began to be seriously discussed at the start of the computer age: Moreover, if we move in the direction of making machines which learn and whose behavior is modified by experience, we must face the fact that every degree of independence we give the machine is a degree of possible defiance of our wishes. In 1988 Blay Whitby published a book outlining the need for AI to be developed along ethical and socially responsible lines. From 2008 to 2009, the Association for the Advancement of Artificial Intelligence (AAAI) commissioned a study to explore and address potential long-term societal influences of AI research and development. The panel was generally skeptical of the radical views expressed by science-fiction authors but agreed that "additional research would be valuable on methods for understanding and verifying the range of behaviors of complex computational systems to minimize unexpected outcomes". In 2011, Roman Yampolskiy introduced the term "AI safety engineering" at the Philosophy and Theory of Artificial Intelligence conference, listing prior failures of AI systems and arguing that "the frequency and seriousness of such events will steadily increase as AIs become more capable". In 2014, philosopher Nick Bostrom published the book Superintelligence: Paths, Dangers, Strategies. He has the opinion that the rise of AGI has the potential to create various societal issues, ranging from the displacement of the workforce by AI, manipulation of political and military structures, to even the possibility of human extinction. His argument that future advanced systems may pose a threat to human existence prompted Elon Musk, Bill Gates, and Stephen Hawking to voice similar concerns. In 2015, dozens of artificial intelligence experts signed an open letter on artificial intelligence calling for research on the societal impacts of AI and outlining concrete directions. To date, the letter has been signed by over 8000 people including Yann LeCun, Shane Legg, Yoshua Bengio, and Stuart Russell. In the same year, a group of academics led by professor Stuart J. Russell founded the Center for Human-Compatible AI at the University of California Berkeley and the Future of Life Institute awarded $6.5 million in grants for research aimed at "ensuring artificial intelligence (AI) remains safe, ethical and beneficial". In 2016, the White House Office of Science and Technology Policy and Carnegie Mellon University announced The Public Workshop on Safety and Control for Artificial Intelligence, which was one of a sequence of four White House workshops aimed at investigating "the advantages and drawbacks" of AI. In the same year, Concrete Problems in AI Safety – one of the first and most influential technical AI Safety agendas – was published. In 2017, the Future of Life Institute sponsored the Asilomar Conference on Beneficial AI, where more than 100 thought leaders formulated principles for beneficial AI including "Race Avoidance: Teams developing AI systems should actively cooperate to avoid corner-cutting on safety standards". In 2018, the DeepMind Safety team outlined AI safety problems in specification, robustness, and assurance. The following year, researchers organized a workshop at ICLR that focused on these problem areas. In 2021, Unsolved Problems in ML Safety was published, outlining research directions in robustness, monitoring, alignment, and systemic safety. In 2023, Rishi Sunak said he wants the United Kingdom to be the "geographical home of global AI safety regulation" and to host the first global summit on AI safety. The AI safety summit took place in November 2023, and focused on the risks of misuse and loss of control associated with frontier AI models. During the summit the intention to create the International Scientific Report on the Safety of Advanced AI was announced. In 2024, The US and UK forged a new partnership on the science of AI safety. The MoU was signed on 1 April 2024 by US commerce secretary Gina Raimondo and UK technology secretary Michelle Donelan to jointly develop advanced AI model testing, following commitments announced at an AI Safety Summit in Bletchley Park in November. In 2025, an international team of 96 experts chaired by Yoshua Bengio published the first International AI Safety Report. The report, commissioned by 30 nations and the United Nations, represents the first global scientific review of potential risks associated with advanced artificial intelligence. It details potential threats stemming from misuse, malfunction, and societal disruption, with the objective of informing policy through evidence-based findings, without providing specific recommendations. == Research focus == AI safety research areas include robustness, monitoring, and alignment. === Robustness === ==== Adversarial robustness ==== AI systems are often vulnerable to adversarial examples or "inputs to machine learning (ML) models that an attacker has intentionally designed to cause the model to make a mistake". For example, in 2013, Szegedy et al. discovered that adding specific imperceptible perturbations to an image could cause it to be misclassified with high confidence. This continues to be an issue with neural networks, though in recent work the perturbations are generally large enough to be perceptible. The image on the right is predicted to be an ostrich after the perturbation is applied. (Left) is a correctly predicted sample, (center) perturbation applied magnified by 10x, (right) adversarial example. Adversarial robustness is often associated with security. Researchers demonstrated that an audio signal could be imperceptibly modified so that speech-to-text systems transcribe it to any message the attacker chooses. Network intrusion and malware detection systems also must be adversarially robust since attackers may design their attacks to fool detectors. Models that represent objectives (reward models) must also be adversarially robust. For example, a reward model might estimate how helpful a text response is and a language model might be trained to maximize this score. Researchers have shown that if a language model is trained for long enough, it will leverage the vulnerabilities of the reward model to achieve a better score and perform worse on the intended task. This issue can be addressed by improving the adversarial robustness of the reward model. More generally, any AI system used to evaluate another AI system must be adversarially robust. This could include monitoring tools, since they could also potentially be tampered with to produce a higher reward. Large language models (LLMs) can be vulnerable to prom
Style sheet (web development)
A web style sheet is a form of separation of content and presentation for web design in which the markup (i.e., HTML or XHTML) of a webpage contains the page's semantic content and structure, but does not define its visual layout (style). Instead, the style is defined in an external style sheet file using a style sheet language such as CSS or XSLT. This design approach is identified as a "separation" because it largely supersedes the antecedent methodology in which a page's markup defined both style and structure. The philosophy underlying this methodology is a specific case of separation of concerns. == Benefits == Separation of style and content has advantages, but has only become practical after improvements in popular web browsers' CSS implementations. === Speed === Overall, users experience of a site utilising style sheets will generally be quicker than sites that do not use the technology. ‘Overall’ as the first page will probably load more slowly – because the style sheet AND the content will need to be transferred. Subsequent pages will load faster because no style information will need to be downloaded – the CSS file will already be in the browser’s cache. === Maintainability === Holding all the presentation styles in one file can reduce the maintenance time and reduces the chance of error, thereby improving presentation consistency. For example, the font color associated with a type of text element may be specified — and therefore easily modified — throughout an entire website simply by changing one short string of characters in a single file. The alternative approach, using styles embedded in each individual page, would require a cumbersome, time consuming, and error-prone edit of every file. === Accessibility === Sites that use CSS with either XHTML or HTML are easier to tweak so that they appear similar in different browsers (Chrome, Internet Explorer, Mozilla Firefox, Opera, Safari, etc.). Sites using CSS "degrade gracefully" in browsers unable to display graphical content, such as Lynx, or those so very old that they cannot use CSS. Browsers ignore CSS that they do not understand, such as CSS 3 statements. This enables a wide variety of user agents to be able to access the content of a site even if they cannot render the style sheet or are not designed with graphical capability in mind. For example, a browser using a refreshable braille display for output could disregard layout information entirely, and the user would still have access to all page content. === Customization === If a page's layout information is stored externally, a user can decide to disable the layout information entirely, leaving the site's bare content still in a readable form. Site authors may also offer multiple style sheets, which can be used to completely change the appearance of the site without altering any of its content. Most modern web browsers also allow the user to define their own style sheet, which can include rules that override the author's layout rules. This allows users, for example, to bold every hyperlink on every page they visit. Browser extensions like Stylish and Stylus have been created to facilitate management of such user style sheets. === Consistency === Because the semantic file contains only the meanings an author intends to convey, the styling of the various elements of the document's content is very consistent. For example, headings, emphasized text, lists and mathematical expressions all receive consistently applied style properties from the external style sheet. Authors need not concern themselves with the style properties at the time of composition. These presentational details can be deferred until the moment of presentation. === Portability === The deferment of presentational details until the time of presentation means that a document can be easily re-purposed for an entirely different presentation medium with merely the application of a new style sheet already prepared for the new medium and consistent with elemental or structural vocabulary of the semantic document. A carefully authored document for a web page can easily be printed to a hard-bound volume complete with headers and footers, page numbers and a generated table of contents simply by applying a new style sheet.