Max Kreminski
LightNews LightNews
banner
maxkreminski.bsky.social
Max Kreminski
@maxkreminski.bsky.social
1.8K followers 380 following 240 posts
https://mkremins.github.io
Posts Replies Media Videos
maxkreminski.bsky.social
🧶 Fuzzy Linkography: Automatic Graphical Summarization of Creative Activity Traces

we introduce a new way to analyze user activity logs from creativity support tools…

…& illuminate patterns of creative divergence, convergence, idea development, & homogenization at scale

arxiv.org/abs/2502.04599
Teaser figure (Figure 1) from the linked paper. Caption: "Fuzzy linkography allows for the rapid translation of user activity logs from digital creativity support tools (and other traces of creative activity) into rough graphical summaries, suitable for visual and quantitative inspection by researchers." On the left, a little sketchy person labeled "CST user" sits at a laptop and types in an idea to be rendered as an image. From their interactions, we derive a "creative activity trace" listing off all the ideas they've suggested as bullet points: "cat", "pink cat", "angry cat", and so on, all the way down to "pink cat with glasses". This trace passes through a process labeled "fuzzy linkography" (depicted here as a fuzzy purple cloud with a cat face and tail), and produces a graph labeled as a "fuzzy linkograph" that shows semantic and temporal relationships between user-suggested ideas. Another little sketchy person, labeled "researcher", peers at the fuzzy linkograph through a magnifying glass and thinks about the patterns they see. Sketches by @cheesetalk.bsky.social!
June 23, 2025 at 8:39 PM
maxkreminski.bsky.social
i've got a ticket open so i think it'll happen *eventually*. the initial email reply did not inspire a great deal of confidence tho
Dear Prof. Max Kreminski, I regard to the inconvenient caused.
April 22, 2025 at 9:41 PM
maxkreminski.bsky.social
extreme alternative to this might be the way that BECOME A GREAT ARTIST deliberately sets you up to “fail” in its Exam Mode: you’re tasked with reproducing a “beautiful” representational image but all the actual affordances just make glitch art. misuse as *only* possible use igf.com/become-great...
A screenshot of BECOME A GREAT ARTIST IN JUST 10 SECONDS. Two images are juxtaposed: a slightly pixelated photo of a landscape, featuring forests and water, on the right, and a glitchy image of bright magenta and lime scanlines on the left. At the top is text reading "10.34% SIMILAR"; at the bottom is text reading "Lesson 3: How to draw a landscape".
April 16, 2025 at 2:17 PM
maxkreminski.bsky.social
good philosophical deep-dive on “value capture”: the phenomenon of people outsourcing their values to external metrics; the mechanisms by which it occurs; the potential benefits and (especially) costs philpapers.org/rec/NGUVCH
Screenshot of linked article's title page: "Value Capture", by Christopher Nguyen. Abstract: "Value capture occurs when an agent’s values are rich and subtle; they enter a social environment that presents simplified — typically quantified — versions of those values; and those simplified articulations come to dominate their practical reasoning. Examples include becoming motivated by FitBit’s step counts, Twitter Likes and Re-tweets, citation rates, ranked lists of best schools, and Grade Point Averages. We are vulnerable to value capture because of the competitive advantage that such crisp and clear expressions of value have in our private reasoning and our public justification. There is, however, a price. In value capture, we take a central component of our autonomy — our ongoing deliberation over the exact articulation of our values — and we outsource it. And the metrics to which we outsource usually engineered for the interests of some external force, like a large-scale institution’s interest in cross-contextual comprehensibility and quick aggregability. That outsourcing cuts off one of the key benefits to personal deliberation. In value capture, we no longer adjust our values and their articulations in light of own rich experience of the world. Our values should often be carefully tailored to our particular selves or our small-scale communities, but in value capture, we buy our values off the rack. In some cases – like decreasing CO2 emissions – the costs of non-tailored values are outweighed by the benefit of precise collective coordination. In other cases, like in our aesthetic lives, they are not. This suggests that we should want different values suited to different scales. We should want value federalism. Some values are perhaps best pursued at the largest-scale level, others at smaller scales. The problem occurs when we exhibit an excess preference for the largest-scale values – when we consistently let the universal metrics swamp our quieter interests."
February 10, 2025 at 7:29 PM
maxkreminski.bsky.social
the a/c is on. they have water and are listening to their favorite music blogs.nasa.gov/crew-10/2024...
Highlighted text from linked NASA press release about astronauts currently on the International Space Station: "The space station recently received two resupply flights in November and is well-stocked with everything the crew needs, including food, water, clothing, and oxygen. The resupply spacecraft also carried special items for the crew to celebrate the holidays aboard the orbital platform."
December 18, 2024 at 10:41 PM
maxkreminski.bsky.social
been looking for a name for this concept for ages!

“endocept”: the vague, pre-articulate sense of vision that guides creation

a core difficulty of creativity support tool design is that in early-stage creativity it’s easy to accidentally displace the endocept

creativity.netslova.ru/Endocept.html
Screenshot of linked dictionary entry: Endocept (from Greek endo, inside), the term introduced by Arieti (1976) meaning a cognitive function which is realized in amorphous cognition; the diffuse and abstract idea often accompanied by strong but unverbalized emotions ('I know what I mean and feel, but I can't explain'). Endoceptual or non-representational states usually have place on the early stages of creative process such as incubation. The endocept is opposite to the concept, a mature form of cognition that can be expressed to other people by the person who experiences or produces it. The endocept has also been referred to by other authors as nonverbal, unconscious, or preconscious cognition.
November 19, 2024 at 4:23 PM
maxkreminski.bsky.social
game narrative people like to talk about how worldbuilding isn’t storytelling. i agree; in fact i think worldbuilding is actually its own emerging creative practice, with its own virtuosos & masterpieces; it overlaps with storytelling but often feels to me more like architecture, hypertext, collage
cover of James Gurney's "Dinotopia" cover of "Dragonology: The Complete Book of Dragons" screenshot of the cool "guys with weapons sneakily floating up a rock" intro scene from Dune Part 2 cover of "Redwall" by Brian Jacques
November 18, 2024 at 7:28 PM
maxkreminski.bsky.social
closely related to a key insight in one of my fav underdiscussed books, "Collaborative Circles": members of historically notable collaborative circles have tended to form especially high-trust pairs & work within these pairs to develop ideas before presenting them to the wider group
Screenshot of a quote from page 23 of Michael P. Farrell's "Collaborative Circles: Friendship Dynamics and Creative Work": Most episodes of creative work occur within these pairs. Although the dynamics leading up to the pairing may include the whole group, it is the paired members who are likely to make the discoveries or to develop the style or the theory that defines the group culture. The moments of discov­ery, the "collaborative moments," take place when a pair of friends are so open and trusting with one another that they can share their wildest, most tentatively held ideas. In these moments, new ideas seem to emerge from the dialogue without "belonging" to either of the pair, and afterward they may not be able to say who had the ideas first. As they share their ideas, experiences, and ways of thinking, the partners become like two linked computers, sharing one another's hard drive mem­ories and software programs. I refer to this type of dyadic exchange as "instrumental intimacy." While engaged in this type of exchange, members are vulnerable to one another. When they open up and exchange wild ideas that go far beyond accepted approaches in their field, they not only risk appear­ing sloppy, foolish, or even mad; they risk having someone steal their most original ideas. Before two members are able to establish this kind of exchange, they usually must test one another's responses in less intimate exchanges and negotiate a high degree of trust. In an escalating set of exchanges, the collaborative pair develop a set of norms that facilitates trust, openness, and exchange.
November 18, 2024 at 1:26 AM
maxkreminski.bsky.social
paper of the day: virtuosic human creators (unlike modern creative ML systems) often succeed by "amplifying the anomaly" – making a weird but distinctive choice & repeatedly revisiting it within a single work, transforming it from "wrong" to "right" via development mkremins.github.io/refs/Amplify...
Abstract of the linked paper: Both large language models (LLMs) and the human brain develop internal models of reality to make accurate predictions. Both typically prefer choices with the strongest track records. However, when faced with a creative challenge, LLMs remain committed to high-probability options while humans can opt for unproven ones. This paper delves into one way of making unlikely events plausible—“amplifying the anomaly.” The concept involves extrapolating viable consequences from an unlikely proposition. Rather than being treated as oddball events or “one-offs,” the anomaly permeates the creative work. Notably, novelty and appropriateness can be in tension with each other, with high utility coming at the cost of low novelty. Amplifying the anomaly aligns these competing demands. It enhances originality: the rarer a proposition and the more thoroughly it is worked out, the more unique and surprising the result. At the same time, the effectiveness and value of the unproven option also rises: thorough elaboration in the creative product establishes its fitness. Musical examples by Beethoven, Schubert, and contemporary composer Sky Macklay, along with creative products from other domains, illustrate this principle. Classic LLMs have several limitations that make it difficult to amplify the anomaly: they are steered toward norm-driven outcomes, make short-term decisions, and are not designed to self-evaluate. As a result, they have difficulty of developing unusual propositions in a non-obvious way without human guidance. Alternatives approaches, including adversarial networks and team AI, are briefly examined. Implications for the future of computational and human creativity are discussed.
November 16, 2024 at 9:59 PM
maxkreminski.bsky.social
(maybe) most known for: the time i got mad about people shit-talking Unity games & accidentally wrote the approachable-creative-tools manifesto that still motivates my research today mkremins.github.io/trashifesto/
Screenshot of the linked website. Visible text: Trashifesto by Max Kreminski On July 19, 2017 I was mad about people denigrating novice-friendly creative tools. In response I posted a thread of tweets. This thread is probably my single most impactful piece of writing to date; it's also maybe still the most accurate existing description of why I do what I do. It now seems that Twitter has more or less drawn to a close, so I reproduce the tweets in question below. a reminder: if inexperienced creators are using your tool to churn out loads of half-baked garbage, your tool is a phenomenal success [...]
November 13, 2024 at 6:30 PM
maxkreminski.bsky.social
we also support an alternative "destroy meaning" mode (we called this "waluigi mode" during development) that tries to make minimal edits maximally destructive to meaning

this occasionally yields some great asemic writing
An example of Savelost's "destroy meaning" mode at work. Given the initial lyrics "such distance from our friends / like a scratch across the lens", Savelost reduces it to surprisingly pronouncable nonsense: "su isefm ou fed/leatrosstheln".
January 5, 2024 at 12:42 AM
maxkreminski.bsky.social
🫠 new poetry machine: Savelost

given a sentence, Savelost repeatedly picks one letter to delete…

while trying to keep the new sentence as semantically similar as possible to the orig sentence

concept + algo by me, js port + frontend by @barrettrees.bsky.social

barrettrees.com/savelost
A screenshot of Savelost in action. The initial sentence "What isn't saved (will be lost)" is iteratively reduced, character by character, to the name of the project: "Savelost". Full text of the depicted poem: What isn't saved(will be lost) What isn't saved(will be lost What isn' saved(will be lost What isn'saved(will be lost What is'saved(will be lost Wat is'saved(will be lost Wat s'saved(will be lost Wats'saved(will be lost Was'saved(will be lost Was'save(will be lost as'save(will be lost a'save(will be lost 'save(will be lost 'savewill be lost 'savewil be lost 'savewi be lost 'savew be lost 'save be lost save be lost savebe lost savee lost save lost savelost A longer example Savelost output, beginning with the lyrics "Do you believe what you're saying? Yeah right now but not that often!" and eventually reducing them to the single letter "t". Along the way, Savelost produces partially reduced sentences like "you believe what you'saying?Ye often!" and "believe?often".
January 5, 2024 at 12:41 AM
maxkreminski.bsky.social
still think about this constantly www.nytimes.com/2018/10/25/m...
Impressed by Latour's dedication, Gaillardet remarked that Latour could have been a scientist. The idea seemed almost too much for him to bear. "I could have been a scientist," Latour said with arch gravity. "I've wasted my life." "Oh, Bru-no!" Gaillardet said, in the way in which one might comfort a wounded bird. "To produce one fact!" Latour sighed, and pointed a finger in the air, as though to demonstrate its indisputable solidity. There was an ache in his eyes. "Can you imagine the pleasure of producing one fact?"
December 6, 2023 at 10:56 PM
maxkreminski.bsky.social
🔠 Blabrecs: An AI-Based Game of Nonsense Word Creation

longer writeup of my AI-based wordgame Blabrecs, which i'll be showing off (w/@isaac_karth) as part of the NeurIPS Creative AI exhibition this year

lmk if you'll be at NeurIPS and want to meet up!

mkremins.github.io/publications...
Visualization of how the original Blabrecs classifier makes a decision about the plausibility of two five-letter nonsense words: TEOSS and ZOIBN. Each word is split into a series of three-letter trigrams and the probability of each trigram is judged according to a statistical model of English dictionary words. Then the probabilities of these trigrams are multiplied together to yield a combined probability for the word as a whole, which is compared to the average probability of all same-length dictionary words. The combined probability of TEOSS is higher than the average probability of all five-letter dictionary words, so TEOSS is allowed to be played; ZOIBN is too unlikely, so it's judged as a fake word and excluded from play.
November 28, 2023 at 8:56 PM
maxkreminski.bsky.social
there’s maybe something here about great art falling into the audience’s zone of proximal development: between what they fully *can* predict (cliché zone, filter stays active) and what they fully *can’t* predict (noise zone, filter reactivates)
Three nested concentric circles representing the theory of the "zone of proximal development" from educational psychology. The innermost circle is labeled "Learner can do unaided"; the middle circle is labeled "Learner can do with guidance"; the outermost circle is labeled "Learner cannot do". The ZPD (zone of proximal development) is the middle circle; it contains skills and knowledge that the learner can be successfully assisted in developing based on what they already know.
October 24, 2023 at 11:01 PM
maxkreminski.bsky.social
tip for retrieval-augmented generation research: steal ideas from the old literature on case-based reasoning! CBR ≈ symbolic RAG

open-domain human/AI collaborative storytelling isn’t unique to modern LLMs; Say Anything was doing it with CBR back in 2008

ict.usc.edu/pubs/Say%20A...
Figure 2 from the linked paper: An example of a story, titled car chase, created with the system and given a coherence rating of 4. On the left are the sentences the user submitted to the system. In the middle are the highest ranked sentences in the database and on the right are the next sentences returned to the user by the system.
September 10, 2023 at 8:33 PM
maxkreminski.bsky.social
playin the hits
Three-panel comic modified from Nedroid's "I made this". First panel: a guy labeled "AI-based creative tool" hands a weird object to a guy labeled "human user". Tool: "You made this." User: "I made this?" Second panel: user looks down at the object, an ambiguous look on their face. Third panel: user's expression becomes contented. User: "AI made this."
August 17, 2023 at 11:51 PM
maxkreminski.bsky.social
🙊 Toward Better Gossip Simulation in Emergent Narrative Systems

describes a four-phase abstract approach to simulating gossip (witness → reflect → propagate → decay) + a toy impl of this approach

to be presented next week @ CoG!

https://mkremins.github.io/publications/Gossamer_CoG2023.pdf
Abstract: Interactive emergent narrative games often make use of social simulation techniques, including the modeling of character relationships and knowledge, to generate compelling gameplay and stories. Existing attempts at simulating the spread of knowledge and opinions between characters, however, strictly limit the informational and social content of the communications that characters exchange, making it difficult to create play experiences in which gossip is a core mechanic. In this paper, we introduce gossip simulation as an open problem in intelligent narrative technologies and present an abstract approach to informationally and socially rich gossip simulation for emergent narrative-oriented social simulations, as well as a preliminary concrete implementation of this approach. Index Terms: procedural content generation, emergent narrative, social simulation, story sifting
August 16, 2023 at 6:26 PM