446 Matching Annotations
  1. Last 7 days
    1. collection of data traditionally considered unprotected—such as public movements—draws questions about whether that information deserves heightened legal protection when collected at a mass scale.

      on scale

    1. Their relationships to the platforms facilitatingthese transactions was sometimes contentious, given that workerswere not officially employed by marketplace platforms, but thelatter had a high impact on the workflows of laborers.

      digital labor spacemaking

    2. “Trash Worker” is a large umbrellaof occupations and levels of agency, which includes human indi-viduals and corporations, but also mold, bacteria, fungi and insects(a controversial theme among the community of Trash Workers).

      human-bot contestations?

    1. It defers the decision about whether to serve a visual puzzle to a later point in the flow after more information is available from the browser.

      is this not what noCAPTCHA reCAPTCHA does? let you in if norm, if not then it presents the image-based reCAPTCHA

    1. Lin, in his contribution, traces the politicsof provision in a sector whose technological and organizational coordinates are changingrapidly: airline food.

      Fresh Not Frozen connections

    2. development of integrated transnational supply chains hasenabled capital to exploit differences among workforces in different parts of the world,creating new regimes of labor containment and fragmentation based on ostensibly noneco-nomic features of identity (race, ethnicity, nationality, citizenship status, etc.)

      东华金龙?wayfair conspiracy?

    3. In their drive to quantify and optimize circulation, logisticalimaginaries can only enact themselves through the production of space, thereby suturinga form of calculative reason premised on system-wide optimization to the reconfiguration ofphysical and social landscapes.

      wdtm

    1. Naor’s hunch implies an eventual changing of guards: if sufficiently auto-mated, the computer itself becomes the gatekeeper, no longer the mere administratorof a database already compiled and refined by humans. What’s more, Naor takes careto recommend that, if this automated identification process is to remain secure in theface of adversaries, it should make publicly available the program that is used to gen-erate each test. The implication here—a profound one—is that security models thatdepend upon the withholding of key information are ultimately much less durablethan models that prey on the ostensible differences in human and nonhuman inter-pretative capacity.

      re: self,

      A major tension I realized within this line of reasoning is that it demands more “transparency” and insight, context and information about CAPTCHA, which is tough to contend with since these tests serve as mundane infrastructure for web security that depends on a certain level of mystique and enigma. Does security require a certain performance of impenetrability in order to work? Castle walls of yore, fortifications work because they are brute stone that block unwanted visitors—but do they also work for their architectural... “aura?” This accordingly leads me to ponder various “generative AI,” “AI agent”-flavored questions that might disrupt or upset present assumptions of CAPTCHA... (i.e., that it is a human-bot difference test—assumptions of the test’s purpose and how it should be administered...)

    2. one’s identity functionally reduced to the ongoing productionof identifiable content.

      yep:

      This positions personhood and humanity as a convenient and helpful ideological and emotional framework, skeuomorph, and metaphor through which to conceptualize the hcomp system involved; “humanness” is defined negatively as “not displaying bot-like patterns” and recursively defined through successful interactions with the system, which are all by necessity opaque for security reasons, presumably.

    3. Rather, it sought to productively convene multiple unwitting internet users, bring-ing them into contingent relation in order to identify content vis-a-vis consensus.

      yep, what I wrote about a community of workers who are on call 24/7 on demand, and yet atomized and individualized: an totalizingly isolating (?) experience

    4. rather was deemed to be accurate inasmuchas it manifested an index of social consensus

      how you are judged — as a data point in relation to/comparison with all the other fellow human data labelers?

    5. . Has von Ahn inadvertently furnished a critical insight long bandied aboutin science and technology studies, or does his decentering of the human point towarda fraught sociopolitical precipice?

      the tension that I have identified, in this reading, then is that we are operating in a relational model whilst the motivations are stuck in the realist

      re: second part of question, exo influence lingers and my note on CAPTCHAs not as proving we are human, but that we are somehow still needed / desired participants ... in some capital exchange scheme/system? deep fried degraded eroded human participation online?

    6. it is both the wellspring out of which CAPTCHA’srelationalparadigm emerged,and a bellwether of the“deep learning”revolution in artificial intelligence that wouldcrest over the subsequent decade, itself a relational alternative to the realist traditionof“symbolic AI.”

      i.e., DL as "learn from the data" — namely that produced by hcomp?

    7. (The essentialism of perceptual faculties accorded to different types of users is yetanother indication of the realist foundation underlying this approach.

      i.e., not thinking about accessibility / assumption of a "normative" user's faculties?

    8. Thisrealistversusrelationaldistinction calls to mind a long lineageof humanistic scholarship, with particularly deep roots in science, technology, and infra-structure studies, but takes its foremost inspiration from Johanna Drucker’s critical rework-ing of the aesthetic foundations of data visualization practices (Drucker,2011,2014).1

      Vardouli Graph Vision too?

    9. The heuristic triad I have extractedfrom Benjamin’s assessment of the unsteady relation between humans andmachines—viz.“fleeting and secret images,”“the associative mechanism,”“smaller andsmaller”—shapes and reflects this article’s attempt to historicize the peculiar, yet illus-trative curio of internet history that is CAPTCHA.

      holy shit?? resonant???

    10. an affirmation that thecamera is capable ofsui generisperceptive operations, catching glimpse of“fleeting andsecret images”which elude the human interlocutor;

      cf Flusser, technical images? (Flusser on gesture?)

      Farocki on operational image

      data based TD art popularity

    1. In the perverse world of coastal engineering the project is billed as an example of “building with nature,” when in reality it is an essential part of that country’s never-ending war with nature.

      "building"

    1. analyse

      in what sense are they analyzing? beyond mathematically based (ontologically speaking (? did I get this right)) mental models of "meaning"?

  2. Oct 2025
    1. a conscientious engagement with the politics of citation that is mindful of how citational practices can be tools for either the reification of, or resistance to, unethical hierarchies of knowledge.

      notions of prestige, too. how evaluation/values higher/further? up/downstream affect citation choices

      citing chicly. putting yourself in conversation with a certain group — implicitly defined aforementioned "citation cartels?"

    1. why are women in science less productive than men, in almost all academic disciplines and regardless of the productivity measure used?

      wdtm: productive?

  3. Sep 2025
    1. Fallout is therefore understood primarily retro-spectively but lived in the future anterior—a form of history made visible innegative outcomes.

      future anterior: lived in the pasts of the future

    1. Semantic zooming is a technique thathas been used to address this. It suggests visualizing different levelsof details at various levels of zooming scales [8].

      i.e. "different levels of scale/read"

    2. heyconcluded that complex and interdependent creative tasks, such aspaper and fiction writing, thrive in human-AI co-creation due tothe iterative and interdependent nature of the interactions

      check sponsors

    3. This collaborative style of work hasalready found application in numerous domains, such as musiccomposition [46], creative coding [39], writing [27,40,68,69,71],design [38,42,44,47], and video authoring [63].

      CHECK CITES

    1. LLMs, such asGPT-4,possess an extensive range of encoded knowledge, yet they mightlack domain-specific information for specialized or emerging topics,as well as for topics involving confidential or sensitive information.
    2. we presented the tool as an AI-powered oracle that could potentially be fallible or overlook certainfactors and encouraged users to conduct their own explorations inaddition to relying on Selenite’s insights.
    1. 7 F1 points

      would it be realistic to expect contextual threads / menus / popups / windows as SG had explained to RN with me in attendance in the spring?

    2. adaptive personalization of the highlights (i

      how often does this become Windows Recall like? does this mean something like Windows Recall? what are you capturing and how?

    3. readers often felt uncomfortable when theysaw long, unhighlighted passages where they thought important information likely could be found.

      expectation v how system/tech interprets expectations...?

    4. extractive andabstractive methods for generating summaries from long-form documents have been developed over the years

      wdtm "extractive and abstractive"? == methods to summarize?

    5. fluid documents that provided contextual access to supplementalinformation between lines of text [14], fluid hypertext [88], visualizations for social annotations within papers [28],and affordances for annotating papers and jumping readers to passages of interest [25,68].

      cf for proj

    6. These studies reveal a tension between reader expectationsand system design, because it is not always possible to highlight according to passage importance while achievinga desirable distribution of highlights. Readers also desired some influence over the quantity and distribution ofhighlights within a paper

      personally i would almost prefer the main points that authors bold.

      acknowledged on p3.

    7. studies on the effect of social annotation on attention within public multimediacontent [13], news reading [39], and education [24,87,90].

      cf for project

      "popular highlights" seem to drive popularity bias. something about being able to go through the text at least once yourself, then as you take notes throughout v. at the end, having the "social context" help probe your thinking?

    8. Without sophisticated controls and affordances enabling more goal-driven or personalized skimming, highlightsonly present a single pathway through a paper

      will it necessarily be the first hierarchical marker that people turn towards? what do these additional considerations really suggest, and what would a meaningful implementation of them achieve?

    9. The tools could also lead a reader to pay less attention to the paper as they skim, once they are no longerrequired to drive the skimming process themselves.

      what might this mean for how the reading comprehension process plays out?

    1. importance of studying howcollaborativedecision-making—particularly in real-world organizational settings,in the presence of existing interpersonal and power relations amongdecision-makers—impacts how people rely upon and make senseof AI models.
    1. It is because all these inscriptions can be superimposed, reshuffled, recombined, and summarized, and that totally new phenomena emerge, hidden from the other people from whom all these inscriptions have been exacted.
    2. But there is a major drawback with ideograms; once gathered you cannot array them in a cascade in such a way that thousands of records can be turned in one, that is literally “punctualized” through geometrical or mathematical skills.

      hm!, ?

    3. A bureau is, in many ways, and more and more every year, a small laboratory in which many elements can be connected together just because their scale and nature has been averaged out:

      hm?

    4. Booker retraces the history of engineering drawings (1982). Linear perspective (see above) progressively “changed the concept of pictures from being just representation to that of their being projections onto planes” (p. 31).
    5. A more powerful theory, we submit, is one that with fewer elements and fewer and simpler transformations makes it possible to get at every other theory (past and future).

      举一反三ability

    6. Most of what we impute to connections in the mind may be explained by this reshuffling of inscriptions that all have the same “optical consistency”. The same is true of what we call “metaphor” (see a funny case in Woolf, 1975; see also Latour and Woolgar, 1979: chap. 4; Goody, 1977; Hughes, 1979; Ong, 1982).

      hm

    7. n politics as in science, when someone is said to “master” a question or to “dominate” a subject, you should normally look for the flat surface that enables mastery (a map, a list, a file, a census, the wall of a gallery, a card-index, a repertory); and you will find it.

      thank you

    8. The superimposition of the first peak with the slope in the physiograph starts to produce an objectwhose limits are the visual inscriptions produced in the lab. The object is a real object no more and no less than any other,

      world within a world

    9. No, says Eisenstein, he is the first notto look at the sky, but to look simultaneously to all the former predictions andhis own, written down together in the same form.

      (compu) TFT machine/brain go brr

    10. Just because we are used to this setting, and breathe it like fresh air, does not mean that we should not describe all the little innovations that make it the most powerful device to achieve power.

      instruments that measure measuring instruments. see (there it is again!) prev1 and prev2

    11. the cost of dissenting increases with each new collection, each new labelling, cach new redrawing.

      you might as well register your own product/trademark/patent; you might as well found your own firm? I feel I am being reductive in the latter part here.

    12. She argues, and rightly so, that an image, a diagram, cannot convince anyone, both because there are always many interpretations possible, and, above all, because the diagram does not force the dissenter to look at it. She sees the interest in inscription devices as an exaggeration of the power of semiotics (and a French one at that!). But such a position misses the point of my argument. It is precisely because the dissenter can always escape and try out another interpretation, that so much energy and time is devoted by scientists to cornerhim and surround him with ever more dramatic visual effects.

      !!!

    13. a trend in these cascades. They always move on the direction of the greater merging of figures, numbers and letters, merging greatly facilitated by their homogeneous treatment as binary units in and by computers.

      wdtm in 2025?

    14. as first pictures, then a statistical rendering of the events; but with an increasing competition for the construction of harder facts, the articles now include more and more layers of graphic display, and the cascade of columns summarized by tables, diagrams, and equations is still unfolding.

      verisimilitude; 身临其境

    15. ?

      my whole process during the summer tryna figure out what the "atomic unit of data vis/rep" is for Cahill; then realizing hm is my mental model out of whack?

    16. Collections are essential but only while the archives are well-kept, the labels are in place, and the specimens do not decay. Even this is not enough, since a museum collection is still too much for one “mind” to handle. So the collection will be drawn, written, recoded, and this process will take place as long as more combinable geometrized forms have not been obtained from the specimens (continuing the process through which the specimens had been extracted from their contexts).

      speculative information/data mining; chatting with archive folks during summer

    17. Every time there is a dispute, great pains are taken to find, or sometimes to invent, a new instrument of visualization, which will enhance the image, accelerate the readings, and, as Lynch has shown, conspire with the visual characteristics of the things that lend themselves to diagrams on paper (

      hm!

    18. Scientists start seeing something once they stop looking at nature and look exclusively and obsessively at prints and flat inscriptions11. In the debates around perception, what is always forgotten is this simple drift from watching confusing three-dimensional objects, to inspecting two-dimensional images which have been made less confusing.

      Cahill

    19. Bourdieu’s critique of ethnography (1972) is that once this first violence has been committed, no matter what we do, we will not understand the savages any more. Fabian however, sees this mobilization of all savages in a few lands through collection, mapping, list making,archives, linguistics, etc. assomething evil. With candor, he wishes to find another way to “know” the savages.

      hm!

    20. to emergeas fullfledged sciences from their carefully kept files. The “panopticon” is another way of obtaining the “optical consistency” necessary for power on a large scale.

      --> threatening / dangling visualization.

    21. when a visual vocabulary is invented that replaces the manipulations by calculation of formulas.

      languages that formalize entire disciplines. shared rituals. astro "pretty pictures". IP and patents. shape grammars.

    22. . The privilege of the printing press comes from its ability to help many innovations to act at once, but it is only one innovation among the many that help toanswer this simplest of all questions: how to dominate on a large scale?

      dude,

    23. In this, the “proof race” is similar to the arms race because the feedback mechanism is the same. Once one competitor starts building up harder facts, the others have to do the same or else submit.
    24. the dissenter will have to do the same thing as his opponent. In order to “doubt back”, so to speak, he will have to write another book, have it printed, and mobilize with copper plates the counterexamples he wants tooppose. The cost of disagreeing will increase8.

      incredibly curious

    25. No irreversible gains could be made, and so no large-scale long-term capitalization was possible.

      historicity; fixed obj in time / explains "if no pictures it didnt happen"; VC?

    26. errors are accurately reproduced and spread with no changes. But corrections are also reproduced fast, cheaply and with no further changes. So, at the end, the accuracy shifts from the medium to the message, from the printed book to the context with which it establishes a two-way connection. A new interest in “Truth” does not come from a new vision, but from the same old vision applying itself to new visible objects that mobilize space and time differently7.
    27. !Impossible palaces can be drawn realistically, but it is also possible to draw!possible objects as if they were utopian ones.

      that which can/must not be represented

      My Name is Red

    28. The speakers!are talking to one another, feeling, hearing and touching each other, butthey are !now talking withmany absent things presented all at once.

      what is an alternative method that explicates the baggage of cartesianism?

    1. ausersearchingforgluten-freeitemsinasupermarketmayneedtorepeatedlyask,“Isthisitemgluten-free?”andlistentoresponses,“No,thesenoodlesarenotgluten-free.”

      depends? in what ways specifically are "AI assistants" being used?

    1. “Do Cellular Phones Dream of Civil War? The Mystification of Production and the Consequences of Technology Fetishism in the Eastern Congo,” in Inclusion and Exclusion in the Global Arena, ed. Max Kirsch and June Nash (Routledge, 2006), 78
    1. 17

      shapes are mutually defining and read as a single unit whether...

      19

      the figure-field relationship ceases to be as such, and all shapes interact with one another.

      33 consider when a pattern can be strongly directional — how does it read when turned across diff angles?

      34, on game C, Manipulations

      move from simple positions of the design unit to more complex by carefully controlling the repeat size, its dark-light balance, and its alignment with itself

      Figure 1-35 (p37)...!!!

    1. the landscape of media consumption is set up now so you never have to do the equivalent of eating your vegetables, and that means you never get to suddenly realize that you love Brussels sprouts.”

      nothing in moderation, always in excess

    1. Each of these transformationsspawns (or generates) new centers, which fit in a natural way into the system of centers already in existence, and supportsthe wholeness which was there initially.

      a meme: a person sitting in a pond, pool of jade water surrounded by plants. image is overlaid with the text, 'Unbothered. Moisturized. Happy. In My Lane. Focused. Flourishing. '

    1. Visualization of the 256x256 Hilbert curve by making pixels brighter the later they are visited by the curve.

      is this "tiled over" / applied to each pixel? can we think of it sorta like a diffusion matrix? or is it "stamped" over / in blocks?

    2. To quantize the current pixel, the last nnn quantization errors are added to the current pixel with weights given in the diffusion sequence. In the article they use an exponential falloff for the weights — the previous pixel’s quantization error getting a weight of 1, the oldest quantization error in the list a small, chosen weight rrr.

      wdtm??

    3. Limiting the number of pixels a single pixel can influence together with the organic look

      "organic look" via hilbert curve construction/curviness/走向? and limiting the number of pixels influenced because... going by the curve's order, instead of a matrix?

    4. Atkinson Dithering.

      patterns in water make me wonder if theres a way to programmatically "reverse engineer" "source" images from the specific types of tinier patterns you get from dithered?

      i.e., dithered image --> choose favorite dith algo --> get original image; to see what forms/figures/compositions yield specific patterns when diffused? this feels way too broad / undirected though.

    5. it doesn’t diffuse the entire error to neighboring pixels, increasing the perceived contrast of the image.

      how are you choosing normalization constants / scalars?

    6. error diffusion algorithms that we won’t touch on in this post is that they can handle arbitrary color palettes, while ordered dithering requires your color palette to be evenly spaced.

      bc... you are changing image as you go along?

    7. convolution (which is the underlying operation of a Gaussian blur) has to loop over each field of the Gaussian kernel for each pixel in the image. However, if you convert both the image as well as the Gaussian kernel to the frequency domain (using one of the many Fast Fourier Transform algorithms), convolution becomes an element-wise multiplication.

      cf RMO compression studies? sublimation was it?

    8. Afterwards, every pixel gets a number between 0 and n (where n is the total number of pixels) according to their importance for forming clusters and voids.

      why?

    9. deterministic and parallelizable per pixel

      wdtm in computing context: "deterministic"? determined by a single formula?

      assuming parallelizable ≈ something like multitaskable?

    10. equivalent:

      how? via some garbly ChatGPT:

      zero-mean jitter to the brightness and compare to a fixed midpoint (0.5). Zero-mean is important: it means the randomness doesn’t systematically push pixels lighter or darker. It only adds uncertainty around the threshold.

    1. with virtual teams trust does notprogress as it does with traditional teams, inwhich different types of trust emerge in stages;instead, all types of trust emerge at the beginningof the relationship

      to compensate for not being physically together?

    1. With Bentham’s plan for prison architecture, we can see how light, shad-ows, mirrors, and walls are all employed in ways that are meant to engen-der in many a prisoner a certain self- discipline under the threat of external observation, as was its intended function.

      cf spectale opera seeing being seen theatre, set des

    2. “No matter how different, or even opposite the purpose: whether it be t

      infra: Silver/KE? on modernity? modularity and buildings as moldable assets

    3. Pastoral power is a power that is individualizing, beneficent, and “essentially exercised over a multi-plicity in movement.”5

      sprawling hills

    1. Hemmenttetal.callforartistic,designerlypracticesofrevealingthe“distor-tionsinthewaysinwhichalgorithmsmakesenseoftheworld”[29];yetdonotoutlinehowthismayrelatetoexistingdesignre-searchmethodologiesaswellasactualMLtechnologies.
    2. RedströmandWiltsefurtherinterrogatehowuserinteractionsaretiedtoinfrastructuralfunctions,andoutlinehow“surface-levelsimplicity”ofinteractionssuchaspressingplayinSpotifybelie“dynamic,sophisticated,andhiddenbackendcomplexity”[61].

      use for Rich infra class

    3. integrativeprototypingmethodsrefectingboth“MLstatisticalintelligenceandhumancommonsenseintelligence”[10]aremissinginthefeld.

      Ken's augi project?

    4. a“regions-of-error”techniqueshowingthemodeluncertaintyofpredictions[30].Similarly,Kinkeldeyetal.usealandscapemetaphorinaclustervisualization,indicatingthroughagrey-scaletopographyhowcertaintheclusteringmodelisaboutthemembershipofeachindividualpointbytheirlocationin“peaksorslopes”[36].

      how does the buzzword (so to speak..) of "latent space" figure here?

    1. Some additional dependencies are required to access all example datasets in skimage.data. Install them using: python -m pip install -U scikit-image[data]

      unable/couldnt do this..

    1. has yet to ever produce a single manifesto

      lack of staking claim in the embeddedness of infrastructure — tfw your job is the most politicized and yet you claim apoliticalness?

    1. Based on this information, theteacherthenmadearichinferenceaboutthelatent,under-lyingcauseof the behavior, and responded with supportand flexibility that an AI tutor could not provide

      never taking it at face value; opening up opportunities for conversation and understanding

    1. The general rationale behind the idea of Hybrid Intelli-gence is that humans and computers have complementarycapabilities that can be combined to augment each other.

      capabilities are defined in relation to each other.

    2. Ensuring interpretability and transparency of machinelearning models while maintaining accuracy

      definitions: interpretability, transparency, accuracy

    1. Here physicists and social scientists are using network theories and algorithms to model, mine, and understand these processes.

      what does this approach gain? what does it obscure/hide/lose?