1. Last 7 days
    1. “tea” and “sea” would be farther apart because they have dissimilar meanings and are not used together often, even though they have similar spelling.

      They are not used often together.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. Types of vector embeddings

      What type of embedding we have in our vector database: word, sentence or document. Could we have embedding of all three?

      ||JovanNj|| ||anjadjATdiplomacy.edu||

    Created with Sketch. Visit annotations in context

    Created with Sketch. Tags

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. Computing Power and the Governance of Artificial Intelligence

      @jovan @sorina This article (published just a week ago) was a new research on regulating frontier AI models by imposing restrictions and safety measures on AI chips.

      I previously shared a similar article called 'How to Catch a Chinchilla', which expresses preliminary ideas of this one.

      The article is focused on regulating the 'hardware' part of frontier models both because it's more visible and easier to track/locate and because, as far as the current way AI is built are concerned, development of advanced AI models still rely on securing a good number of AI chips.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Tags

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. By that text, titled “Promotion of inclusive and effective international tax cooperation at the United Nations” (document A/C.2/78/L.18/Rev.1), the Assembly would stress that efforts in international tax cooperation should be universal in approach and scope and fully consider the different needs and capacities of all States, in particular developing countries and countries in special situations.  The Committee approved it by a recorded vote of 125 in favour to 48 against, with 9 abstentions (Armenia, Costa Rica, El Salvador, Iceland, Mexico, Norway, Peru, Türkiye, United Arab Emirates).

      Vote on drafting UN Taxation Treaty.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. Following the UN Secretary-General’s report earlier this year highlighting the exclusionary nature of the OECD’s tax work, however, we now see a most welcome formal challenge.

      Interesting document

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. by despite the tendency for multi-stakeholder processes to water things down”

      ||sorina|| it is interesting aspect that multi-stakeholder processes are used to water things down.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. I'm struggling with getting to the group page. every time I log in I land on the textus page.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

  2. Feb 2024
    1. one common definition of weight is the importance given to individual input features when training a machine learning model

      What is exactly weight? There are two explanatons in this paragraph. ||JovanNj||||anjadjATdiplomacy.edu||

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. Factors such as the intense US-China tech rivalry and the escalating chip embargo on Chinese AI firms further diminish the government’s incentive to impose strict regulation.

      ||MariliaM|| This article discusses AI governance in China. It has linkages between digital and security as you are focusing on.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. distributing power, catalyzing innovation, and ensuring transparency

      Three reasons for open source AI.

    2. n the EU, open foundation models trained with fewer than 1025floating point operations (a measure of the amount of compute expended) appear to be exempted under the recently negotiated AI Act.

      EU AI Act on open source foundational models.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. For each key finding below, more granular response categories are presented in the main text, along with demographic breakdowns of interest.

      Why America was afraid of AI in Spring 2023.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. LangChain and LlamaIndex.

      Are we using any of these methods?

    2. a time-weighted approach would help us retrieve the most recent document.

      Can we use this for updates, courses, blog posts?

    3. we use list comprehension to get text embeddings for all text chunks.

      ||anjadjATdiplomacy.edu|| does it mean that we get at the end one vectoral value or whole text (not only text chunk).?

    4. it is crucial to split the document into smaller chunks so that it's more effective to identify and retrieve the most relevant information in the retrieval process later.

      Can we split it into smaller chanks? Ideally it should be on the level of sentence or paragraph in the text. Can we use some of HTML markers/code to identify beinning/end of para or sentence.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. The Act takes a “risk approach”, i.e. it conditions intervention on the basis of the expected “risk” of the application – but not a market power approach.

      EU AI Act ignores market power risks.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. as the Federal Government, aim to coordinate our international digital policy even more closely with our partners

      Continuation of block-thinking, divides and paralysis.

    2. value-based technology partnerships

      This says it all about partnerships. Either you are like-minded and democratic, or, partnerships will be limited.

    3. same time, we are encouraging the commitment and representation of like-minded partner countries of the Global South in relevant Internet governance forums as well as the reciprocal capacity building and expansion

      More participation and capacity building. BUT, for like-minded.

    4. We use digitalization to tackle global challenges

      Although the layout masks it a bit, much more attention and space were given to resilience than to sustainability.

    5. The high complexity of global value and supply chains requires close collaboration and specialization at EU level and with partners and allies worldwide

      Friendshoring

    6. EU to connect additional partners to existing infrastructures and make space-based data, services and products accessible to them.

      EU Global Gateways?

    7. We reject any political influence that runs counter to our values

      I though standards were a technical-only area! Just being ironic...

    8. We also take into account at all times the increasing security and defence policy dimensions of data policy

      Securitization of data flows. ||MariliaM||

    9. The deteriorating geopolitical environment increases the urgency of constantly considering also other dimensions, particularly security and defence policy, when dealing with regulatory issue

      This creates the link to coordinate with EU Economic security strategy, and with EU and German China strategies.

    10. o promoting digital skills along the entire education chain

      Focus on education.

    11. to counteract the possible leakage of critical technologies and knowledge.

      This is 'Chinese norm'

    12. with our values and interests.

      A strong focus on 'our' in values and interests. It remains to be seen if 'others' relate to those who have different values and itnerests (but not confrontational) or only to those who directly oppose 'our' values and interests. It will be key distinction in this 'value-based' approach. Does it include cooperation with China?

    13. improving geo-redundancy.

      Interesting new concept. What would it mean in practice?

    14. the expansion and protection of submarine data cables, terrestrial optical fibres and new satellite constellations

      Critical infrastructure.

    15. democratic partners

      is there any indication of criteria (e.g. elections, free press)? Is there any list of those partners?

    16. also

      It opens the space for other format such as multilateral (unlike previously mentioned digital networking).

    17. innovative, secure, trustworthy, human-centric and sustainable artificial intelligence (AI).

      to follow this phrase in AI debates. It could emerge as 5-pillars slogan for AI governance.

    18. in particular through our cooperation in ‘Team Europe’.

      Does it mean that cooperation with Africa, and the neighbours will be more pushed via EU-channels, including 'Team Europe'?

    19. we are enhancing bilateral cooperation with partners who share our values and who are important political, economic and regulatory players in the digital sphere.

      Is this AND or OR condition between two elements: - 'share our values' - important players

      For example, does China fit into this criterioon?

    20. We consider the multi-stakeholder approach essential for global digital networking.

      What is 'global digital networking'? Is it cybersecurity as well? What about AI?

    21. We want to avoid duplicate structures, which make a broad-based and inclusive participation of stakeholders in international bodies harder

      Highly relevant link between duplucate structire as a way that prevent inclusive participation.

    22. economic interdependencies in the area of digitalization do not adversely affect our security.

      An interesting interplay between interdependence and security with a possible shift closer to security.

    23. Especially in the Global South, standards can be a way to remove obstacles to development, promote innovations and stimulate economic growth in a sustainable manner.

      It is the far-fetched link. Standards can be useful. But, developing countries have so many other obstacles (access, funding) apart from standards.

    24. We advocate for corresponding binding rules on digital trade at international level. Moreover, we support the efforts of the EU with regard to rules on digital trade, in particular in trade agreements. We are coordinating closely with the other EU Member States and the European Commission in international negotiations. We also take into account the challenges faced by the countries of the Global South.

      Interesting! There is no mention of WTO e-commerce negotiations explicitly which was a pillar of Germany's e-commerce and data policies. Does Germany follow US shift away from data negotiations in WTO? It can be followed by statements during WTO Ministerial conference.

    25. he FAIR principle (findable, accessible, interoperable, reusable) for data.

      to be further analysed

    26. the concept of Data Free Flow with Trust (DFFT)

      to be further analysed

    27. We are promoting broad-based options for participation in order to make digital policy decision-making processes and digital participation as inclusive as possible.

      A possible support of Diplo?

    28. This includes our commitment in the G7, the Group of Twenty (G20), the Organisation for Economic Co-operation and Development (OECD), the Global Partnership on Artificial Intelligence (GPAI), the Council of Europe, the World Trade Organization (WTO), international financial institutions such as the World Bank and the UN, including the ITU. We coordinate closely with the other EU Member States and the European Commission.

      Key global partners of Germany.

    29. net neutrality is an overriding principle in the Internet ecosystem.

      Almost forgotten concept.

    30. or a global, open, free and secure Internet

      We should monitor how many countries use this 4-pillar phrase in their strategies, speeches, etc.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. Russia also ranked 38th out 193 countries in the latest AI -readiness index by Oxford Insights, a consultancy; America came first.

      ||sorina|| ||Jovan||

    2. Yandex, Russia’s search giant, has integrated an LLM , Yandex GPT -2, into its virtual-assistant service, known as “Alice”.

      ||JovanNj|| Alice - Russian GPt. Do they have API?

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. Many in Europe hope the DSA will ensure major social media companies are held more accountable for how they protect elections from disinformation.

      Test

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. Another Hanna paper, presented at the Resistance AI workshop, urges the machine learning community to go beyond scale when considering how to address systemic social issues and asserts that resistance to scale thinking is needed.

      From what I observed, the notion of 'scaling an algorithm' always results in high decontextualization of data and outputs (i.e., training data is taken out of the context from which it was derived, and the outputs are applied to various contexts instead of merely the one the model was trained for).

      And a huge problem for the lack of reflections in this regard was, in my opinion, the tech community's blatant refusal to think about the systemic, societal risks of their models--they framed it as the problem of policymakers and ethicists, and that the social can be separately mitigated apart from the technical conception of the models.

    2. “We argue that fixes that focus narrowly on improving datasets by making them more representative or more challenging might miss the more general point raised by these critiques, and we’ll be trapped in a game of dataset whack-a-mole rather than making progress, so long as notions of ‘progress’ are largely defined by performance on datasets,” the paper reads. “Should this come to pass, we predict that machine learning as a field will be better positioned to understand how its technology impacts people and to design solutions that work with fidelity and equity in their deployment contexts.”

      I suspected that the current AI regulations discussions are missing out on long-term risks that certain paradigms in the current AI development space might impose; one is the predominant idea that we can resolve fairness by simply codifying fairness metrics and making the latter a requirement of a model to 'hit the market'.

      This logic fails to challenge what this paragraph is trying to say here: ensuring fairness goes beyond the idea of getting a 'fair dataset' that performs well; it concerns what the deployment context is, how developers and users are to interact with the AI model, and how does the model become interwoven in the lives of the people working around it.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Tags

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. as well as clarifying the principles and norms under which various organizations should operate.

      Deos it mean that it will instruct ITU, UNESCO about princples and norms?

    2. A ‘distributed-CERN’ reimagined for AI, networked across diverse states and regions, could expand opportunities for greater involvement

      CERN is already distributed.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. Chaos theory is about systems where small changes to the initial conditions result in extremely large changes in the results.

      Problem with predicitons

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. Sometimes referred to as data-centre alley, northern Virginia is home to just over three square kilometres of data centres, most of which are within 75 square kilometres in Loudoun County.

      ||sorina|| Here is relevan paragraph for you.

    2. Because AI is based on matrix maths, it involves large blocks of computation being done at once, which means a lot of transistors have to change states very quickly. That draws a lot more power than normal computer tasks, which flip far fewer transistors at once for a typical calculation.

      Why AI takes more energy?

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. The GDPR protects the personal data of individuals in the EU. Personal data is defined as: “any information relating to an identified or identifiable natural person (‘data subject’).”

      Milos testing

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. However, it warned the cables at some points run at a depth of 100 metres, reducing the need for hi-tech submarines. In 2013, three divers were arrested in Egypt for attempting to cut an undersea cable near the port of Alexandria that provides much of the internet capacity between Europe and Egypt.

      This is a problem with cables ||sorina||

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. for the first time

      ||AndrijanaG||||sorina|| Was it adopted for the first time?

    2. its own Global AI Governance Initiative in October, calling on major powers to take “a prudent and responsible attitude” on military use of artificial intelligence technologies.

      Do we have t

    3. the Political Declaration on the Responsible Military Use of Artificial Intelligence and Autonomy a year ago.

      ||sorina||||VladaR||||MariliaM|| Do we have anything on this declartion?

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. A paper by Rebecca Johnson, a researcher at the University of Sydney, published in 2022, found that Chat GPT -3 gave replies on topics such as gun control and refugee policy that aligned most with the values displayed by Americans in the World Values Survey, a global questionnaire of public opinion.

      To be checked

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. Firms in countries that have not signed on to the American export-control regime, like Singapore, can buy chips and send them on to Chinese entities without the knowledge of the American firms or the Department of Commerce.

      Singapore does not join American export restrictions to China.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. Publishers around the world are all too aware of this shift; over half of those recently surveyed by the Reuters Institute for the Study of Journalism said that they plan to devote more effort into putting stories on TikTok this year.

      @jovan Major shift

    2. “It has to be short, it has to be fast,”

      It is useul guidelines for DW ||Jovan||

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. The emphasis on assessing academic performance by how many papers a researcher can publish, for example, acts as a powerful incentive for fraud at worst, and for gaming the system at best.

      This could question whole academic system - push for more publishing.

    2. Checking models against reality is what science is supposed to be about, after all.

      Good analogy with role of science in human society.

    3. That can cause “model collapse”.

      Model colapse can be triggered by using AI-genereated data for development of the model. ||JovanNj||||anjadjATdiplomacy.edu||||sorina||

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. the company dominates the market for AI accelerator chips, accounting for 86% of such components sold globally;

      86% of AI chips are produced by Nvidia

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. The concept intentionally obfuscates (clouds, one might say) the user’s ability to see the existence of hardware.

      Cloud i

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. Taylor Swift, the latest high-profile victim of a deepfake, might disagree.

      H

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. let’s get rid of bloody nationalism, let’s eradicate corruption, let’s stop the brain drain, let’s provide a decent life and, above all, a future to all our citizens, let’s take care of nature and let’s live in peace. The rest – EU, UN etc. – may come sooner or it may come later: what matters most is that we take better care of people. 

      Very good point! But, elites and mafia cannot make money in such scenario. It is easier to divide people and take money and resources from them.

    2. Bioregionalism, thus the rivers (White Drin, Lepenac, Ibar) and their plains and basins on all sides, could function as an excellent ‘negotiator’ and connect efficiently Kosovo to Albania, Macedonia and Serbia without mobilizing the narrative of ‘Great Serbia’, ‘Great Albania’, etc.

      Great proposal. Unfortunately, it is uitopian. But, worth keeping alive.

    3. The reality check is mind-blowing: the end of SFR Yugoslavia (1991-1992), the end of FR Yugoslavia (Serbia and Montenegro) in 2006, the loss of Kosovo (2008) and, let’s not forget, a ‘brain drain’ of unprecedent magnitude. Except for the potential breakup of the United Kingdom (Northern Ireland and Scotland), no other European country has faced such devastating consequences of its policy failures.

      Agree!

    4. the EU is quite supportive of Serbia... as it needs a strong player in the region.

      It is for debate. I agree that it is the case. But, I think it is more 'tactical' than 'strategic' reasoning. EU does not need conflict in the Balkans and they want to close 'Kosovo chapter'. They do not see Serbia as long-term partner.

    5. Both Vučić and Bosnian Serb leader Milorad Dodik have invested so much in pro-Russian sentiment that they are unable to wean themselves off it.

      True!

    6. What is noteworthy is that Serbia is increasingly being mentioned in the same way.

      I am not sure it is the case. Serbia is more mentioned as 'Western Balkans'.

    7. Consequently, the natural European anchorage for the post-Yugoslav states is Central Europe – naturally, I am viewing this in the framework of the European Union.

      Interesting 'shift of geography'.

    8. the Western powers are not in control of this new conflictual world where power has become powerless, while weakness has given rise to power to the point of destabilizing the agenda of the strongest.

      There is a good point here.

    9. Politics is driven less by state initiatives and more by social dynamics –

      I am not sure that it is the case. It could be the case on deeper level. But, we are facing exactly opposite - power polticis everywhere.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

  3. Jan 2024
    1. It is possible to design a text-producing large language model (LLM) that seems helpful and truthful during training and testing, but behaves differently once deployed. And according to a study shared this month on arXiv1, attempts to detect and remove such two-faced behaviour are often useless — and can even make the models better at hiding their true nature.The finding that trying to retrain deceptive LLMs can make the situation worse “was something that was particularly surprising to us … and potentially scary”, says co-author Evan Hubinger, a computer scientist at Anthropic, an AI start-up company in San Francisco, California.

      Kind of scary indeed. It takes us back to the question of trust in developers.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Tags

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. four-fifths of grade six maths teachers in South Africa did not understand the concepts they were supposed to teach.

      What about training them?

    2. A chatbot can give undivided attention to each child, at any time of day, and never gets tired

      Makes sens

    3. The third reason is that developing countries have gaping shortages of skilled workers: there are nowhere near enough teachers, doctors, engineers or managers.

      Should we use AI or education of people? I can understand using AI in Europe with labour shortage, but in Africa, it does not make sense at all ||sorina||

    4. There are three main reasons for optimism.

      These arguments are not related only to Arica. They are general. Can they apply to Africa? ||sorina||

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. Some developers in India are already taking Western models and fine-tuning them with local data to provide a whizzy language-translation service, avoiding the heavy capital costs of model-building.

      b

    2. the phone in their pockets.

      Ok

    3. Pupils in Kenya will soon be asking a chatbot questions about their homework, and the chatbot will be tweaking and improving its lessons in response.

      Is it g

    4. The imf says that a fifth to a quarter of workers there are most exposed to replacement, compared with a third in rich countries.

      Interest

    5. Because emerging countries have fewer white-collar workers, the disruption and the gain to existing firms may be smaller than in the West.

      Possible tru

    6. Most exciting of all, it could help income levels catch up with those in the rich world.

      How?

    7. As it spreads, the technology could raise productivity and shrink gaps in human capital faster than many before it.

      Interesti

    8. N ew technology brings with it both the sweet hope of greater prosperity and the cruel fear of missing out.

      Opportunitiy/threat binary.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. The boss has to show their face to employees regularly, and it cannot be the face of someone who looks like they haven’t slept for two weeks. They have to glad-hand the board, meet investors, attend endless networking events and make time for actual work. It is exhausting to contemplate, let alone

      ||sorina|| why it is not good to be boss.

    Created with Sketch. Visit annotations in context

    Created with Sketch. Annotators

    Created with Sketch. URL

    1. with national

      What about 'international standardisation organisations'.

    2. as the WTO, dispute resolution can also be facilitated through global forums.

      Dispute resolution failed in the WTO. Again, we have governments as actors.

    3. 72. Reporting frameworks can be inspired by existing practices of the IAEA for mutual reassurance on nuclear safety and nuclear security, as well as the WHO on disease surveillance.

      In both examples, reporting is done by member states. How realistic is to use this in the case of AI. Is analogy useful?

    4. a techno-prudential model, akin to the macro-prudential framework used to increase resilience in central banking

      How useful is this analogy between AI which is very diverse and rather centralised system of central banks.

    5. The possibility of rogue AI escaping control and posing still larger risks cannot be ruled out.

      Extinction risk stenence

    6. A new mechanism (or mechanisms) is required to facilitate access to data, compute, and talent

      New structuers/mechanisms

    7. The borderless nature of AI tools

      What is 'borderless' nature? AI is created and used within certain jurisdictions always.

    8. could also be coordinated through a body that harmonises policies, builds common understandings, surfaces best practices, supports implementation and promotes peer-to-peer learning

      new body

    9. A consensus on the direction and pace of AI technologies

      What type of consensus is expected?