1. Apr 2022
    1. Four countries—America, China, India and Russia—have conducted such tests, most recently Russia in November last year.
    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. China’s Xuzhou Construction Machinery Group (XCMG) operates an industrial digital platform in Germany that focuses on technology transfer and talent training for its operations in Germany and in China.
    2. hese platforms facilitate the transfer of advanced manufacturing capabilities and talent training from leaders in the field such as Germany
    3. Foreign digital platforms that compete and partner with China include Microsoft’s Azure, PTC’s ThingWorx, Siemens’ MindSphere, and SAP’s HANA Cloud Platform.
    4. In the developing world, it is working with countries that lack energy access and may assess that current energy distribution systems are unfair.
    5. In the developed world, GEIDCO is appealing to environmental interests in clean energies and efficiencies.
    6. To develop the platform from the technology infrastructure side, State Grid and China’s other state monopoly in the power sector, China Southern Power Grid, are working with Huawei’s global energy business unit to use its data and cloud services. Huawei provides these firms cloud-based data services for their smart-grid operations in China.
    7. CEPRI also oversees the development and operation of China’s domestic digital platform for electric power allocation and trading.
    8. he Global Energy Interconnection and Development Cooperation Organization (GEIDCO),
    9. The plan promotes Chinese renewable technologies and products (wind, solar, and hydropower) and State Grid’s ultra-high voltage AC/DC hybrid-power grid that it can overlay on traditional grids to transmit renewable energy.
    10. tate Grid,
    11. Si m i la rly, China’s national blockchain technology could be used within these digital payments, which could allow greater interoperability and also points of visibility and control for the state
    12. In January 2022, Chinese Android and Apple app stores offered a pilot digital currency “e-CNY” app that was developed by PBOC’s Digital Currency Institute for use in trial cities and locations in China hosting the Olympics.
    13. seek to gain advantages in standardizing the technologies and systems it is using that could in turn be adopted by other countries.
    14. to issue a sovereign digital currency as part of efforts to internationalize the renminbi, reduce dependence on the global U.S. dollar system, and safeguard China’s monetary sovereignty
    15. t the Bank of International Settlements (BIS) Innovation Summit in March 2021, China submitted a proposal on global digital governance that discusses its views for standards and norms on cross-border digital transactions, risk supervision, and the use and ownership of data.

      ||sorina|| An interesting standardisation initiative.

    16. Alibaba almost certainly gains data and insights from the manufacturing and logistics firms that use its platform.
    17. Ahead of the October 2021 G-20 Leaders’ Summit, Xi Jinping announced China’s intent to join the Digital Economy Partnership Agreement (DEPA), an agreement launched by Singapore, New Zealand, and Chile
    18. The Zhejiang FTZ plans promote the global role of eWTP and Zhejiang’s role as one of two national agriculture and energy stockpiling bases.
    19. The Chinese government initially proposed the concept for eWTP at the 2016 G-20 Leaders’ Summit held in Hangzhou.
    20. China’s digital industrial platforms are partnered with advanced manufacturing leaders in Germany to gain competencies
    21. to diversify away from its dependence on the U.S. dollar and promote alternative payment systems with other countries
    22. n trade, Alibaba’s eWTP (electronic World Trade Platform) is leading the Chinese government’s efforts to create digital trade zones with partner countries and working to build out China’s global logistics capabilities to support this platform.
    23. to secure greater access to global data, increase its control over information flows in third countries, shape the emerging digital ecosystems in developing countries, and ultimately exert greater influence over the global digital domain.
    24. The Trump administration’s efforts to ban TikTok and WeChat from the U.S. market due to national security concerns faced pushback in part because of how such a restriction might curtail cross-border communication options.
    25. Most international policy tools leveraged thus far to protect against China’s digital platforms and the information superiority they might provide the Chinese government are tactical in nature and consider risks on a case-by-case basis.
    26. hese digital platforms are also breaking into sensitive and otherwise highly regulated foreign markets that are considered critical infrastructure, including finance, health, media, public utilities, telecommunications, and government services.
    27. China’s policies restrict foreign access to its digital market and increasingly curtail the ability of foreign firms to collect or leverage data from China.
    28. Data is increasingly the underlying source of value in economic and geopolitical activity.
    29. Its platforms may appear to be open, but that openness tends to be limited to areas where China seeks to obtain certain foreign expertise.
    30. China’s OpenAtom Foundation—a Chinese alternative to the Linux Foundation
    31. Huawei is hosting its operating system, HarmonyOS, on Gitee in an effort to expand globally despite U.S. export controls.
    32. Gitee
    33. The Chinese government’s industrial plans for information technology and software development promote the use of open-source technology and its global community of expertise to advance innovation within an “open, collaborative, and international ecosystem.
    34. China is advancing interoperability and common infrastructure and standards that, while often used in global markets to promote commercial development,
    35. a framework blockchain standard project.

      ||sorina|| Do we have anything on this proposal?

    36. many companies are developing and deploying blockchain technologies in applications that promote decentralization and anonymity and attempt to circumvent government oversight.

      Blockchain does not need to be anonymous.

    37. China is aiming to issue a national blockchain standard in 2022.
    38. The Hainan Free Trade Zone is China’s first blockchain pilot zone. In a nod to state priorities, its plan is called Secure, Sharing, Compliance+ (SSC+).54 At a Hainan trade forum in December 2019, participating countries signed a digital agreement with blockchain provisions.
    39. in financial payments, supply chain management, and e-governance.
    40. Measures since 2020 related to antitrust, privacy, and data appear designed to ensure that the state has access to its digital platforms and that these platforms open their proprietary services to one another to facilitate innovation and development of the sector.
    41. China’s recent actions against its technology champions appear aimed at state centralization and convergence.
    42. This use of both private and state-owned companies to increase Chinese influence and control over international digital ecosystems is a core pillar of Beijing’s digital strategy.
    43. n 2018, for example, the Chinese government integrated the national ID card with Tencent’s QR code and is now leveraging Tencent facial recognition technology to enforce its rules restricting minors’ access to gaming domestically.
    44. Convergence aligns with and fuels centralized government control over platforms and their information. It may also make China’s platforms uniquely competitive and powerful.
    45. In Russia, Alibaba’s AliExpress created a joint venture with MegaFun, Mail.ru, and Russia’s sovereign wealth fund. Alibaba has acquired Trendyol (Turkey), Daraz (Pakistan), and Lazada (Singapore) and invested in other firms in India and Indonesia.
    46. ByteDance’s acquisition of Musical.ly in November 2017 allowed TikTok to immediately add an estimated 80 million U.S. users to its platform
    47. Tencent has developed its global gaming network largely through investments in foreign gaming firms, including Supercell and Epic Games.
    48. are also expanding overseas through foreign partnerships and investments.
    49. Many of China’s industrial and innovation platforms use foreign platforms, such as Microsoft’s GitHub, for cross-border technology transfer and collaboration.
    50. Baidu’s autonomous driving platform, Apollo, is based on open-source Android software and operates on Microsoft’s Azure Cloud.
    51. Alibaba, Huawei, and Tencent are examples of platform providers that have rapidly expanded their global cloud services.
    52. any social media and super apps, such as TikTok and WeChat, use a model that allows their platforms to reside within operating systems on mobile phones, such as Apple’s iOS and Google’s Android.
    53. BRI leverages the provision of digital infrastructure and systems, as well as preferential financing, to bypass normal trade liberalization reciprocity rules and allow China to expand in foreign markets without having to open its own market in return.
    54. This asymmetry has allowed Chinese firms to develop within China in a protected market and then expand globally, while U.S. digital platforms are kept out of the world’s largest and fastest-growing market.
    55. Due to the size of China’s market and its focus on mobile networks and digitalization, in absolute terms China’s domestic market represents around 40% of global e-commerce.
    56. Chinese firms are able to grow in the protected, insulated, and government-bolstered Chinese market with little international competition and the advantages of building a globally significant position by leveraging China’s enormous domestic user base.
    57. Until 2012, Chinese technology firms generally pursued a copycat strategy in which they created digital platforms for Chinese users that mimicked leading U.S. digital platforms. Alibaba began in 1999 as a competitor to Amazon, and in 2003 it developed Alipay as a competitor to PayPal. Baidu started as a competitor to Google in search engine and mapping services and developed a competitor video and movie service (iQIYI) and Chinese version of Wikipedia (Baike). Sina Weibo began as a challenger to Twitter. In 2011, Tencent launched WeChat to challenge Facebook’s WhatsApp. In 2012, ByteDance began as a news aggregator, Toutiao, which used an algorithm the company refined in 2016 to launch a short video business, Douyin (the domestic version of TikTok), to challenge Facebook and YouTube.

      Survey of Chinese copycat companies.

    58. China’s digital platforms both advance and build off this global construction of digital infrastructure, including transnational land and submarine cable networks, fiber-optic cables, satellite navigation networks (BeiDou), data centers, and related cloud services.

      to be checked about cables owned by China.

    59. Chinese digital policies outline intentions for China to own and control the intellectual property and standards in global technology value chains as part of broader efforts to gain a leadership position.
    60. China has committed $1.4 trillion over the next five years to digital infrastructure, including in 5G and 6G, smart cities, and IoT applications for manufacturing.
    61. China’s Internet Plus strategy,
    62. While they also have their own commercial interests, Chinese technology firms and the digital platforms they develop and operate are closely tied to the Chinese state.
    63. hina’s new data security law restricts cross-border data transfers, and the new personal information privacy law enhances the state’s authority over the collection and use of data
    64. Since 2020, the Chinese government has strengthened its control over data, algorithms, and digital platform operations.
    65. China still depends on certain foreign technology, hardware, and software to operate its platforms—including sensors, semiconductors, and high-end programmable logic controllers
    66. The Chinese government has also tightened the ability to use workarounds such as VPNs.1
    67. Beijing seeks to create alternative global digital platforms and related architecture that are centered on and controlled by China.
    68. digital platforms are by nature disruptive to established businesses and sectors
    69. China is developing digital platforms to promote Chinese technology and economic competitiveness, enhance political and social controls, and challenge current global trade, technology, energy, information, and financial networks by creating alternatives that it controls.
    70. According to Shiji, over 56% of all hotels in the United States now use its technology in their stack.
    71. A prominent example of a PRC firm operating in this area is China’s state-tied Shiji Group, which operates back-office software systems for the hospitality industry.
    72. through partnerships with leaders in advanced manufacturing such as Germany,
    73. Tencent’s WeChat
    74. hey offer points of connection, exchange, and control.
    75. they are generally understood to be internet-connected and software-based digital spaces that facilitate the exchange of information and the creation of value through the online interactions of businesses and individuals.
    76. China’s digital strategy leverages one-sided market protections and access. Its leading digital platforms—such as Alibaba, Tencent, JD.com, DiDi, and TikTok—have a significant global market share as a result of their dominant position in China’s massive domestic market.
    77. by other governments that have sought to use antitrust authorities and other regulatory approaches to encourage innovation and competition.

      China has strong anti-trust regulation as Alibaba experienced recently.

    78. In absolute terms, China’s domestic market represents around 40% of global e-commerce.1 It has an estimated 1.3 billion mobile internet users and is a top global market for mobile payments. Between 2015 and 2020, China’s digital economy grew faster than any other market, at an annual rate of almost 17%.
    79. his statist approach keeps China’s domestic digital market walled off from foreign competition and global connections not controlled by China while Chinese firms create and expand China’s digital platforms offshore.
    80. China’s digital platforms seek to operationalize its development efforts across the entire technology value chain in hardware, software, and related design, manufacturing, infrastructure, and services.
    81. Digital platforms play a key role in this effort.
    82. are controlled or influenced by China; China’s technology firms have a leadership position; and China governs a significant share of international information flows
    83. These persistent asymmetries due to digital protectionism and state controls allow China to secure a global market position that could become increasingly difficult and costly to counter over time.
    84. to access, analyze, and leverage wide swaths of global data across a range of platforms and applications
    85. China seeks to operationalize its technology development efforts across the entire value chain in hardware, software, and related design, manufacturing, infrastructure, and services
    86. he Chinese government’s global ambitions as conveyed through its Belt and Road Initiative and related Digital Silk Road plans and seek to leverage and integrate the hard and soft infrastructure that China’s firms have established or acquired overseas
    87. China’s digital platforms are the likely place where the full range of China’s technology, economic, and geopolitical efforts, if successful, could converge and solidify China’s position in global markets.
    88. Developing responses that better account for uncertainties around a technology’s trajectory or a country’s ability to translate concepts into capabilities.
    89. Taking a more multidisciplinary approach to due diligence on decisions related to digital infrastructure.
    90. Recalibrating data security policy and privacy frameworks to account for the fact that PRC regulations on each are not motivated by the same drivers as in liberal democracies.
    91. decision-makers must bear in mind that digital technology is constantly evolving.
    92. China’s export of digital infrastructure abroad is a critical element of its broader digital strategy.
    93. eople’s Bank of China officials conduct monitoring using big-data analytics that flag unusual activity that might indicate illegal activity (as defined in the PRC). The People’s Bank of China might also seek to more closely monitor a specific subset of individuals and entities who are targets of the regime.
    94. CEP is already integrated with digital payment technology such as Alipay, which is accepted globally
    95. digital currency electronic payment (DCEP)
    96. he real focus should be on the implications of DCEP as a financial technology.
    97. The standardization also embeds a values system that ultimately runs against liberal democratic values
    98. Chinese technology companies are comparable to U.S. companies in that they tend to occupy all layers of the “technology stack”: the physical layer, the network layer, and the application layer.

      Three layers structure

    99. For example, one GTCOM product, Language Box, was reportedly integrated into a Huawei smart conferencing solution sold as part of smart-city packages.
    100. Global Tone Communications Technology (GTCOM) is a company controlled by the Central Propaganda Department that engages in global big-data collection.
    101. This does not mean that Hisense will use the data for purposes beyond business, but it does mean that the company has the ability to do so.
    102. One such company, Hisense, is a leading global company that sells smart products at more affordable prices than its competitors (using the Roku TV interface).
    103. In the United States, this data has been sold to political campaigns.
    104. For China, there is strategic value in having access to data from which it can extract valuable information to maintain an internationally competitive edge.
    105. These devices ultimately allow for greater visibility and transparency about the state of global supply chains.
    106. The IIoT refers to “the billions of industrial devices—anything from the machines in a factory to the engines inside an aeroplane—that are filled with sensors, connected to wireless networks, and gathering and sharing data.”
    107. The vendor then “can be subject to adverse extrajudicial direction, or the vendor’s poor cyber security posture means they are subject to adverse external interference.”
    108. The risk of a country allowing a China-based company to supply or build a data center is equivalent to allowing a high-risk vendor to build a country’s 5G network.
    109. Data centers are another common type of infrastructure Chinese companies have won contracts for or have built overseas.
    110. Within Serbia there have been concerns about the system being used to target the incumbent government’s political opponents, but the projects persist.
    111. Serbia
    112. Despite—or in some cases because of—the strong link between smart city technologies and coercive state activity in China, these solutions have been exported globally.
    113. In addition to surveillance cameras, Sharp Eyes is focused on building video/image/information exchange and sharing platforms and county-village-township comprehensive management centers.
    114. Skynet refers to video monitoring equipment that is mostly used at major intersections, law-and-order checkpoints, and other public assembly locations
    115. Smart cities, or safe cities, are another example of a project that PRC companies (most notably Huawei) are contracted to deliver overseas.
    116. The data derived from these Chinese-provided technologies overseas becomes part of the party-state’s data ecosystem.

      Is it linked?

    117. If the issue is framed as “a good technology can be misused,” rather than “digital and data-driven technologies can be used to achieve multiple objectives at once, and those uses could be (subjectively) both good and bad,” then the major game-changing effect of the technology in the lens of geopolitical competition is overlooked

      ||sorina|| Do you understand this?

    118. When aggregated, data consisting of seemingly innocuous information can become extremely valuable.
    119. What is overlooked, however, are the fundamental concerns around who has control over systems that enable information flows.
    120. these laws are applicable anywhere in the world where Chinese companies have operations. Even if the companies providing digital infrastructure are acting according to their own market interests rather than by direction of the state, the state can leverage that expansion and market success for its own purposes if and when it chooses.
    121. Article 18 is clear that personal information handlers need not notify individuals that their data is being accessed if other laws and regulations provide that the purpose for that access “be kept confidential or need not be announced.”19
    122. Chinese companies are bound by PRC law no matter the political jurisdiction in which their business operations are located.

      It is like USA Cloud act.

    123. political security is at the core of data security and state security.
    124. The term “central state security leading mechanism” in legal documents is synonymous with the Central State Security Commission (CSSC), which is a CCP agency led by Xi Jinping.
    125. The recently enacted Data Security Law (DSL) and Personal Information Protection Law (PIPL) reinforce these risks. Articles 4 and 5 of the DSL state that the effort to guarantee data security must adhere to the party-state’s “comprehensive state security outlook,” and that the “central state security leading mechanism” is “responsible for decision-making and overall coordination on data security work, and researching, drafting, and guiding the implementation of national data security strategies and relevant major guidelines and policies.”1

      need to be analysed in more details.

    126. not regulating how data passing through a data center or other infrastructure is used.
    127. By controlling the means, one controls the message.
    128. there was an assumption, within liberal democracies in particular, that authoritarianism would be severely undermined and democracy would emerge stronger as improved information flows stymied censorship efforts.
    129. Information security is a prime example
    130. China has recognized, for example, that securing the digital economy, especially as it relates to data security and information security, requires having sufficient control over one’s own digital ecosystem
    131. technology is a tool that helps the party accomplish fusing its political control with China’s economic prosperity and “social development.
    132. digital infrastructure is a key to solving problems in governance and improving its political control.
    133. fiber-optic cables, data centers, and IoT devices enable connectivity in smart cities.
    134. he Internet of Things (IoT), Industrial Internet of Things (IIoT), or other data-dependent environments such as smart cities and smart manufacturing. This infrastructure, better described as “digital infrastructure,”
    135. “new infrastructure,” artificial intelligence, 5G, and data centers.
    136. George Bowden, “MI6 Boss Warns of China ‘Debt Traps and Data Traps,’” BBC, November 30, 2021, https://www.bbc.com/news/uk-59474365.
    137. “If you allow another country to gain access to really critical data about your society, over time that will erode your sovereignty, you no longer have control over that data.
    138. go beyond both surveillance and espionage
    139. in promoting and subsidizing their expansion abroad,

      Is this against WTO rules?

    140. Peter Hartcher, “Huawei? No Way! Why Australia Banned the World’s Biggest Telecoms Firm,” Sydney Morning Herald, May 21, 2021, https://www.smh.com.au/national/huawei-no-way-why-australia-banned-the-world-s-biggest-telecoms-firm-20210503-p57oc9.html; “Huawei 5G Kit Must Be Removed from UK by 2027,” BBC, July 14, 2020, https://www.bbc.com/news/technology-53403793; Gautam Chikermane, “No Huawei in 5G Is a Start, No China in Critical Infrastructure Should Be Next,” Observer Research Foundation, Digital Frontiers, May 5, 2021, https://www.orfonline.org/expert-speak/no-huawei-in-5g-is-a-start-no-china-in-critical-infrastructure-should-be-next; and “Huawei Ban Timeline: Detained CFO Makes Deal with U.S. Justice Department,” CNET, September 30, 2021, https://www.cnet.com/tech/services-and-software/huawei-ban-timeline-detained-cfo-makes-deal-with-us-justice-department

      Discussion on banning Huawei.

    141. to effectively ban Huawei from providing 5G equipment in Australia, the United States, the United Kingdom, India, and numerous other countries, as well as the debate around those decisions.
    142. o violate or exploit globally accepted data privacy norms
    143. he largest issue is that China has a political system that is fundamentally different from liberal democracies and that is embedded in the digital technologies and infrastructure researched and developed in China and exported globally.

      ||sorina|| It is very powerful argument which basically arguest that two technologies cannot co-exist. This argument lifts discussion from technological to value issues which cannot be reconciled by market or governance.

      Tech coexistence will be almost impossible.

    144. embed standards that go against liberal democratic values
    145. he government’s strategic interests
    146. for Chinese digital companies to gain greater market access
    147. access to, and control over, data internationally
    148. export of digital infrastructure
    149. onceived of as agents of geopolitical influence or means to control critical, strategic resources

      Neither has CN been the one starting this trend…

    150. nor delegates to standard-setting bodies have traditionally been
    151. with the ultimate goal of shaping the international architecture rather than simply seeking advantage within it
    152. an edge in setting international standards.
    153. unmatched industrial capacity that allows it to build the physical infrastructure of the digital world.
    154. its centralization allows it to more effectively leverage that scale than any other leading global player.
    155. unmatched size grants it unmatched ability to produce and access data:
    156. but innovation may no longer be the determinative asset it once was
    157. The last industrial revolution, which was catalyzed by the emergence of technology as a factor of production, rewarded innovative capacity as a critical source of national strength.
    158. These conversations ignore the more foundational, strategic contest for the global architecture.
    159. Beijing works to shape the rules of the digital environment from the top down by setting international technical standards and exporting a China-centered system of digital governance.
    160. China’s overarching digital ambition is to seize the opportunity of the digital revolution, control data as a factor of production, become the network great power, and leapfrog to leadership of the world order. This is how Beijing frames the competition for and of the fourth industrial revolution.
    161. mazon’s real power lies in its ability to shape the information ecosystem in which users shop.
    162. it has the dominant information platform.
    163. the player that controls digitized logistics hubs can shape international shipments of cobalt without having to deploy troops to capture mines in the Democratic Republic of the Congo;
    164. power is therefore a function of both capturing data and controlling the architecture of digital exchange: information infrastructure like 5G and smart logistics hubs, platforms like social media and digital trade hubs, and the technical standards and governance systems that define their operations and evolution.
    165. Strategic, revolutionary power in the digital revolution lies a step beyond accessing data—it lies in the ability to shape data and its movement.
    166. owever, data is not like land or labor. Its strategic value does not only come from access or ownership.
    167. China has a first-mover advantage in the digital economy and is expected to achieve a revival in the fourth industrial revolution.4
    168. Technological changes in different periods not only bring about industrial changes, but also affect changes in the world structure.
    169. “With the advent of the digital economy, data elements have become the new engine for economic development. Data is a new production factor, a basic resource, and a strategic resource.
    170. Today’s industrial revolution, the digital revolution, is a function of data having emerged—alongside land, labor, capital, and technology—as a factor of production.
    171. Beijing benefits from a set of asymmetric, structural advantages—scale, centralization, and industrial capacity

      Crucial aspect.

    172. They reveal that China is turning traditionally commercial and cooperative global domains into battlefields of nation-state competition.
    173. China’s digital ambitions threaten the ability of companies to compete fairly in the international marketplace, of information to circulate freely, and of governments to defend themselves.
    174. China intends to define this digital architecture by building its physical infrastructure and corresponding virtual networks and platforms, setting the technical standards that govern them, and shaping the emerging global digital governance regime.
    175. Beijing’s global digital strategy rests on seizing this opportunity by competing to control international data, its movement, and, by extension, the production, distribution, and consumption of resources and ideas internationally.
    Created with Sketch. Visit annotations in context

    Created with Sketch. Tags

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. When we later tune our model to identify the difference between these positive and negative passages, we are teaching it to determine what are often very nuanced differences.
    2. Adding these ‘negative’ training examples (Q, P-) is a common approach used in many bi-encoder fine-tuning methods, including multiple negatives ranking and margin MSE loss (the latter of which we will be using). Using hard negatives in-particular can significantly improve the performance of our models [3].
    3. Excluding the positive passage (if returned), we assume all other returned passages are negatives. We then select one of these negative passages at random to become the negative pair for our query.

      remember

    4. remember

    5. Yes, those returned results are the most similar passages to our query, but they are not the correct passage for our query. We are, in essence, increasing the similarity gap between the correct passage and all other passages, no matter how similar they may be.
    6. It may seem counterintuitive at first. Why would we return the most similar passages and train a model to view these as dissimilar?
    7. Excluding the positive passage (if returned), we assume all other returned passages are negatives. We then select one of these negative passages at random to become the negative pair for our query.
    8. The negative mining process is a retrieval step where, given a query, we return the top_k most similar results.
    9. To fix this, we perform a negative mining step to find highly similar passages to existing P+ passages. As these new passages will be highly similar but not matches to our query Q, our model will need to learn how to distinguish them from genuine matches P+. We refer to these non-matches as negative passages and are written as P-.
    10. The (query, passage) pairs we have now are assumed to be positively similar, written as (Q, P+) where the query is Q, and the positive passage is P+.
    11. Query generation is not perfect. It can generate noisy, sometimes nonsensical queries. And this is where GPL improved upon GenQ. GenQ relies heavily on these synthetic queries being high-quality with little noise. With GPL, this is not the case as the final cross-encoder step labels the similarity of pairs. Meaning dissimilar pairs are likely to be labeled as such. GenQ does not have any such labeling step.
    12. GPL is perfect for scenarios where we have no labeled data. However, it does require a large amount of unstructured text. That could be text data scraped from web pages, PDF documents, etc. The only requirement is that this text data is in-domain, meaning it is relevant to our particular use case.
    13. Each of these steps requires the use of a pre-existing model fine-tuned for each task. The team that introduced GPL also provided models that handle each task. We will discuss these models as we introduce each step and note alternative models where relevant.
    14. Pseudo labeling, using a cross-encoder model to assign similarity scores to pairs.
    15. Negative mining, retrieving similar passages that do not match (negatives).
    16. Query generation, creating queries from passages.
    17. At a high level, GPL consists of three data preparation steps and one fine-tuning step.
    18. As you may have guessed, the same applies to the first scenario of fine-tuning a pretrained model. It can be hard to find relevant, labeled data. With GPL we don’t need to. Unstructured text is all you need.
    19. GPL hopes to solve this problem by allowing us to take existing models and adapt them to new domains using nothing more than unlabeled data. By using unlabeled data we greatly enhance the ease of finding relevant data, all we need is unstructured text.
    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. rights

      It would be ironic that robots get rights in some parts of the world while in other not all humans are considered worthy of having all the human rights. I think first we should work on making sure that ALL humans have rights, and have their basic needs covered and then worry about the machines.

    2. program themselves

      For me it has been complicated to understand how computers can program themselves, or how AI can start to code, or how can AI will achieve singularity - at the end of the day it is the humans that are creating all the coding and the new tech.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. set clear goals for public diplomacy
    2. Clearly defined goals with subordinate country or region-specific objectives where necessaryDesired outcomeDefinition of the problems, risks, and threatsClearly identified target groupsPlan of activities including coordination between relevant national authoritiesDescription of resources and investments requiredStandardised evaluation methodology
    3. can be scaled down based on the needs and resources of each project and ministry.
    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL