Maria Antoniak

@mariaa

NLP, cultural analytics. Assistant Professor of CS at University of Colorado Boulder. Books, bikes, games, art.

1.419 Folgt11.704 Follower3.276 Posts

Bluesky

alt

Ist das dein Profil?Übernimm es — dann gehört dir mariaa.dirty.social.Profil übernehmen

Maria Antoniak@mariaa· 18d

nice reflection from @ninabegus.bsky.social about the limitations and importance of the stories we tell ourselves about robots/aliens/others

For Asimov, however, the three laws were nothing more than a narrative device: a way to generate plot, stage philosophical arguments, and enable his robots’ beautiful failures. Technologists and ethicists often read the laws as something far more consequential: as grand ideas that ought to be imported into AI or neurotechnologies. They assume the laws incorporate a complex human morality ready-made for the machine algorithm—even while surely recognizing that our realities are not just technical but also social and cultural. The latter was precisely Asimov’s message.

In this regard, what gets lost is the other half of his literary project, which was genuinely curious about what it feels like to be something other than human. Less read than his robot stories, and never cited in ethics documents, are his tales from the 1940s and ’50s about alien beings. By his own admission, he was reluctant to write these particular stories to avoid further conflict with his editor at the time, John W. Campbell Jr., who insisted on humans as “superior to other intelligences, even when those others were further advanced technologically.” (Asimov connected this attitude to Campbell’s racist and conservative politics.) Asimov therefore mostly focused on writing stories with “a galaxy populated by human intelligences only,” while steadfastly believing in a universe of multiple intelligences.

Artists, such as Pierre Huyghe and Sougwen Chung, are masters of this exploration. They craft worlds and narratives in which the machine oscillates between operating as an extension of the human body and generating a mystery worthy of further inquiry. Huyghe pushes the boundary between human and nonhuman, masking or hollowing the human face until his figures evade category-dependent description. Chung draws alongside a robotic arm trained on the artist’s own gestures, making the machine a collaborator rather than an instrument.

Maria Antoniak@mariaa· 25d

Screenshot of a Bluesky account, “Argos News” aka render87.bsky.social with a bunch of labels showing that it is posting nonstop without breaks. The bio reads “Global news bot, powered by AI.”

Maria Antoniak@mariaa· 25d

FYI that account is a bot

Labels on the “render87.bsky.social” account showing that the account is posting nonstop without breaks.

Maria Antoniak@mariaa· Jul 9

the weather isn’t cooperating so far, but i did get this view of the tower!

Maria Antoniak@mariaa· Jul 5

ok never mind, you're right, it's annoying. i've added these!

A set of disallowed words and hashtags including "neuro linguistic" and "#neurolinguisticprogramming".

Maria Antoniak@mariaa· Jul 2

Added these:

Maria Antoniak@mariaa· Jul 2

These are the current settings; any match to these is automatically added to the feed. Feedback welcome!

Maria Antoniak@mariaa· Jun 30

not wrong! bambamramfan.github.io/ai-compass/

YOU ARE...
The Kontextmaschine
patron saint: kontextmaschine
It's real, it's powerful, and you find the panic about it as tedious as the hype. You're not interested in whether AI is good or bad — you're interested in who captures the value, who loses their leverage, and which institutions are using it as a story to do what they were already doing. Same as it ever was.

Maria Antoniak@mariaa· Jun 19

New work from my lab! @teagrjohnson.bsky.social built a 12-dimensional narrative framework, annotated Dolma (no small feat given its extreme diversity), and analyzed narrative features across pretraining subsections. Highlight: pretraining data space displays strong narrative organization!

Figure 6: UMAP reduction of SBERT embeddings for 20,000 randomly sampled NARRADOLMA documents, colored by PC1 score (interiority). Labels are based on manual examination. Overlays for all three PCs appear in Fig. A5.

Maria Antoniak@mariaa· Jun 19

A screenshot of my results.

“MARIA ANTONIAK
NLP AND COMPUTATIONAL SOCIAL SCIENCE RESEARCHER
537 STRENGTH · TOP 8%
GPT-5.5 SAYS
COMPUTER SCIENTIST AND NLP RESEARCHER KNOWN FOR WORK ON COMPUTATIONAL SOCIAL SCIENCE AND NATURAL LANGUAGE PROCESSING”

Maria Antoniak@mariaa· Jun 3

This was part of a talk by Alyosha Efros on "Sweet Data." He wanted us all to read this paper, “this is the most important paper in AI”: ieeexplore.ieee.org/abstract/doc... (sorry, much confusion about the speaker's name because they changed the schedule order)

Maria Antoniak@mariaa· Jun 2

One of two matching shop cats at Ophelia’s Books in Seattle.

A fluffy black cat looks at the camera. It’s sitting on a wood floor surrounded by full bookshelves.

Maria Antoniak@mariaa· May 25

good thing i got screenshots 🤷‍♀️ but yeah if you’re engaging with an account and screenshots like this uncritically, maybe reconsider

Maria Antoniak@mariaa· May 22

I'm struggling to find more information about this new #EMNLP2026 policy about "concurrently submitted papers on a related topic with an overlapping set of authors." How related is "related"? How overlapping is "overlapping"? PIs, are you doing this for all your papers in your lab?

Please note the new ARR originality requirement and overlap with other work by the same authors (also can be found in the ARR Call for Papers):

“All submitted papers are expected to describe original, completed and unpublished work. See the publication ethics guidelines on re-use of text from other publications (no more than 10% total tokens). There can be no overlap in stated contributions. ARR now follows ICML in discouraging thinly sliced contributions as follows: any concurrently submitted papers on a related topic with an overlapping set of authors must cite each other and discuss the differences in the related work section. Anonymized versions of such papers should be uploaded as supplementary material. Recommended citation format in the bibliography:

Anonymous (2026). Paper title. Under review.

Papers violating the above originality requirements may be desk rejected.”

In addition, all authors involved may be ineligible to commit their paper(s) to EMNLP 2026 and EMNLP 2027.

Maria Antoniak@mariaa· May 22

done with part iv. now we wait to november 17 (intriguing because our heroine is stuck in november 18) for the next translation. also someone informed me that the translators changed between books ii and iii, and books iii onwards are done by a team of two translators. i had no idea!

The cover of the fourth book in the series. It again is an abstract "heat map" of blues, yellows, and reds, with the heat center inside an oval containing the title and author.

Maria Antoniak@mariaa· May 11

System: You are an expert Named Entity Recognition (NER) system. Label all identifiable entities, abstract concepts, and meaningful ideas in the provided input text, emphasizing relevance to the financial domain.

Ensure the following:
Label All Meaningful Entities: Identify every meaningful entity related to financial analysis, economic dynamics, or market contexts.
Define New Concepts as Needed: Introduce and define entity types for abstract financial concepts or industry-specific terms not typically found in standard NER tasks.
Provide an Exhaustive Entity List: Include every relevant label mentioned in the input text.

Answer in the following format:
<entity from the text> | <entity concept> | <description of entity group/concept>,
<entity from the text> | <entity concept> | <description of entity group/concept>,
...

Here is an Example :
Input:
Lawmakers continue to try to police social media use among teens — but Meta, parent company to Facebook, Instagram, and Threads, is ...

Maria Antoniak@mariaa· May 10

Somehow that got posted without alt text.

It happened one day that an atheist and a man were standing together on a doorstep; and the atheist said, “It is raining.” To which the man replied, “What is raining?” Now the same anarchic mystery that clings round the phrase, “il pleut,” clings round the phrase, “il faut.” In English it is generally represented by the passive mood in grammar, and the Eugenists and their like deal especially in it; they are as passive in their statements as they are active in their experiments. Lady Macbeth said, “Give me the daggers; I’ll let his bowels out.” The Eugenist would say, “In such cases the bowels should, etc.” But when the Eugenists say, “Conditions must be altered” or “Ancestry should be investigated,” or what not, it seems clear that they do not mean that the democracy must do it, whatever else they may mean. The question remains, therefore, whom they do instinctively trust when they say that this or that ought to be done. What is this flying and evanescent authority that vanishes wherever we seek to fix it? Who is the man who is the lost subject that governs the Eugenist’s verb?

Maria Antoniak@mariaa· Apr 20

very cute goodreads alternative

A screenshot of Pagebound showing a navigation bar with search, library, discover, and discuss. underneath are two cards, each featuring some reading activity from another user. The color scheme is pastel and the style is very cute.

Maria Antoniak@mariaa· Apr 13

I started reading “There Is No Antimemetics Division” and while I do very highly recommend it (charming characters, creepy in the best way), I do not recommend reading before bed 😵‍💫

The cover of the book. It shows a misty forest and large stone tower.

Imagine working in an organisation that no one remembers. A division that deals with Unknowns with dangerous memetic properties. There are contagious ideas, which require containment just like any physical threat.The Memetics Division deals with ideas that stick. They literally cannot be forgotten or ignored. They take over the mind of the person who encounters it, like a cult you can’t escape or ignore or be rescued from. This organization finds and isolates those ideas, and prevents their spread.

Maria Antoniak@mariaa· Apr 9

ALSO my sister kindly pointed out that I do this very often with my hands when I'm talking BUT that is kind of how I feel (scheming, pleased) when talking about this experiment!

A famous meme photo of a raccoon with a grin and tapping its fingers together, as if it is plotting something.

Maria Antoniak@mariaa· Apr 6

icon

Screenshots of Eva Stratt (a character in Project Hail Mary) wearing different outfits and making different facial expressions.

Maria Antoniak@mariaa· Apr 2

yep pretty simple! just for our custom blocking as i intentionally have not built in support for larger block lists (for reasons mentioned above)

A screenshot of Lea showing a moderation panel where users can block other users based on who liked or reposted a Bluesky post. The tool warns the user that this list includes people that they follow and that follow them and checks for permission.

Maria Antoniak@mariaa· Mar 31

This email from @colorado.edu about the OpenAI contract is insulting. There was no conversation! There was no shared governance process! Faculty "input" was vocal protest in reaction to a closed-door deal! We still don't have clear information about where the data is stored or who has access to it!

ChatGPT Edu is now available to eligible CU Boulder faculty and staff. This optional, higher‑education–focused generative AI platform is designed to protect academic freedom and privacy, provide equitable access, and offer a more secure alternative to publicly available AI tools.

Student access has been delayed until August 14, following input from faculty governance, as announced in a message from the university president and campus chancellors.
A platform designed for higher education

ChatGPT Edu is provided through a University of Colorado systemwide agreement with OpenAI, ensuring access to a FERPA‑compliant, privacy‑aware environment designed specifically for higher education. Information entered into CU Boulder’s ChatGPT Edu environment is not used to train AI models, helping reduce data security risks associated with consumer tools and better protect student, faculty, staff and university data.
Already using ChatGPT? Make the switch

Generative AI tools are already widely used across campus. Before the OpenAI agreement was signed, more than 28,000 students, faculty and staff created ChatGPT accounts using CU Boulder email addresses, with many more likely using personal accounts. ChatGPT Edu offers a safer, institutionally supported alternative than the publicly available version, providing stronger privacy protections while still delivering the capabilities users expect.

CU Boulder encourages faculty and staff currently using the public version of ChatGPT to consider switching to ChatGPT Edu to benefit from its more data‑secure environment.

Maria Antoniak@mariaa· Mar 31

This is all I see at the link and I’m struggling to find more info. Anyone know more? (To be clear, already plenty to hate in this text, and I have close to zero trust in these companies, but I want to know more about the data.)

We have a commercial relationship with Piazza Technologies Inc. (“Piazza”). Pooja Sankar is the Chief Executive Officer of Piazza and is the spouse of Shyam Sankar, one of our executive officers. During 2017, 2018, 2019, and the six months ended June 30, 2019, we made payments of $412,000, $202,000, $167,000, and $135,000, respectively, for a license to its online recruiting platform and other services provided by Piazza.

Maria Antoniak@mariaa· Mar 28

This was amazing. @kissane.myatproto.social is such an inspiration. Watch the recording if you missed it! Themes: landslides, kelp, social media, democracy, knowledge, agency. Several shoutouts to #atscience! #atmosphereconf

Erin Kissane speaking on the Streamplace live stream. Her slides show an illustration of kelp (by her daughter!) and text saying "Shelter sensemaking: We could build sheltered spaces for community knowledge-making (which would also make a lot of the big-world hostility less necessary)"

Maria Antoniak@mariaa· Mar 26

help

claude prompt asking for prettier colors

claude prompt still asking for better colors

Maria Antoniak@mariaa· Mar 16

I've made it through the first three parts of "On the Calculation of Volume" and only now have discovered that the next four (!) parts are not yet written/translated. They're weird Danish novels about a Belgian woman trapped repeating the same day. Not sure if I recommend but look at these covers.

cover of volume 1: each cover shows an abstract shape with swirling colors

Maria Antoniak@mariaa· Mar 16

yeah these are from the LREC slides he gave at Johns Hopkins in 2004 (worth reading through, they're short!)

Hope Springs Eternal
• My colleagues and I always hoped that linguistics will eventually allow us to strike gold
– HMM tagging (Bahl & Mercer 1976)
• The quote accentuated a certain situation that existed in ASR in the seventies and in NLP in the eighties

The View of the IBM Group
• Linguistic intuition combined with ability to extract information will determine the structure of models and their parameterization
• Parameter values will be estimated from (annotated) data
• We will rely on advice of linguists to create resources
– E.G., Annotating Noun Argument Structure for NomBank
• The problem is not of direct interest to linguists

Conclusions
• Physicists study physical phenomena
- Linguists study language phenomena
• Engineers learned to take advantage of the insights of physicists
• It is our task to figure out how to make use of the insights of linguists

Maria Antoniak@mariaa· Mar 14

I appreciate the intention, but I'm not sure how SocArXiv's new AI policy is enforceable beyond extreme cases and vibes: socopen.org/ai-policy/

We evaluate papers using various procedures:

Moderator assessment of minimal contribution to social science
Identification of author as a human with a consistent identity (as marked by ORCiD or other sources)
Moderator detection of likely-AI slop, with the following red flags:
Literature reviews or synthesis without substantive contribution
Unsubstantiated grand theories, which may have:
Fanciful equations
Integration of multiple major theories into definitive new “paradigms,” “proposed frameworks,” or similar, without substantive engagement
Papers written in many short sections, often including bullet-point lists and numbered paragraphs.
Fake empirical research
Assessment, whether by machine or by moderator perception, of papers with:
High truth-indifference or phoniness (e.g., no realistic grounding)
Enterprise misrepresentation (e.g., false or misleading description of the research project)
Constraint evasion (e.g., implausible or facile assumptions without evidence)
Low falsifiability (e.g., untestable claims without justification)
Absence of methodological accountability (e.g., extensive methods and data discussion without any research output)

Maria Antoniak@mariaa· Mar 8