iiinova — where your digital self keeps evolving

Nick Vincent @nickmvincent.bsky.social · 24/06/2026

Sadly will not be at FAccT this year and will be missing everyone there, but two awesome students from the lab will be there presenting on Social Simulation (dl.acm.org/doi/10.1145/...) and Overreliance dl.acm.org/doi/10.1145/...

dl.acm.org

Mechanism Plausibility in Generative Agent-Based Modeling | Proceedings of the 2026 ACM Conference on Fairness, Accountability, and Transparency

060

Nick Vincent @nickmvincent.bsky.social · 28/05/2026

Yes! Potentially in a disruptive way, however? (Guessing in the ui you’re seeing a bunch all at once?)

110

Nick Vincent @nickmvincent.bsky.social · 07/05/2026

More thoughts on why the surge of interest in AI evaluation, auditing, and benchmarking will create a fresh window to establish healthy data flow and data markets more broadly (largely leaning on old data intermediary / data guild style ideas): dataleverage.substack.com/p/the-ai-eva...

dataleverage.substack.com

The AI "Evaluation Crisis" Is an Opportunity to Get Data Flow Right

Why the AI evaluation crisis could force a reckoning on dataset provenance, attribution, and consent.

110

Nick Vincent @nickmvincent.bsky.social · 03/05/2026

... post -- I was in particular reacting to the use of "augment don't replace" from major figures in the AI industry)

110

Nick Vincent @nickmvincent.bsky.social · 03/05/2026

Really love the entirety of this perspective -- pretty much agree on all fronts. I do think these are exactly the follow up questions that would be critical to ask anyone committing to an "augmentation strategy" (mentioned in other thread, but I didn't do a good job providing context on original...

110

Nick Vincent @nickmvincent.bsky.social · 03/05/2026

In that context, I think there's a bunch of unresolved tension re: the beliefs and assumptions people hold and being very precise about whether this commit will come at the modelling layer, at the data collection layer, in the form you've described above, etc.

110

Nick Vincent @nickmvincent.bsky.social · 03/05/2026

... "augment don't replace" when reacting to societal impact concerns (from journalists, podcasters, etc.). Often while being in an "AGI-believing" social context (i.e. with people who have some belief about continued capabilities scaling, all that can be seen can *eventually* be modelled, etc.)

100

Nick Vincent @nickmvincent.bsky.social · 03/05/2026

Very much appreciate this point! I think this line of thinking is a very viable way to commit. Also realizing that my standalone post was definitely lacking in context making it very straw man-y; this was in response to major figures in AI using some form of...

110

Nick Vincent @nickmvincent.bsky.social · 02/05/2026

Adding this to my schtick-toolbelt of recurring topics to post about; I really think the class of people who aim to shape AI discourse *must* ask: Is a commitment to "augment without replacing" a commitment to intentionally build less capable models or something else?

220

Nick Vincent @nickmvincent.bsky.social · 01/05/2026

(I think it can be achieved by imposing system-level constraints and/or by importing concepts from private and decentralized ML that restrict the flow of data, but I'm skeptical the vision can be achieved at the modelling layer. Please tell me if you think this is wrong!)

020

Nick Vincent @nickmvincent.bsky.social · 01/05/2026

One issue I think that technical AI pundits / the publicly-engaged professoriate / related groups should really push on is trying to get consensus on a technically precise definition of what it would mean to train a model that "augments without replacing"

130

Nick Vincent @nickmvincent.bsky.social · 15/04/2026

Enjoying the “Discussions” session (about online discussions — we’re not just having a “discussion”) in #CHI2026 So far — - conversational voice agents in video calls - visualizing comment threads programs.sigchi.org/chi/2026/pro...

programs.sigchi.org

Conference Programs

000

Reposted by Nick Vincent

Jenn Wortman Vaughan @jennwv.bsky.social · 13/04/2026

At #CHI2026? Check out @sachnishal.bsky.social's talk this Tuesday on how LLM-infused writing tools reshape journalists’ agency — their ability to exercise independent judgment in alignment with their values — in editorial decision making. programs.sigchi.org/chi/2026/pro...

0102

Nick Vincent @nickmvincent.bsky.social · 13/04/2026

Going to #CHI2026 and going to try and engage and boost CHI-related stuff here (and linkedin too, as it seems like there's a lot of activity there)! Hoping to support a positive social media experience around conference-ing, as this was a big part of my positive experience at my first CHI!

0110

Reposted by Nick Vincent

Yuxuan Li @yuxuanli1225.bsky.social · 11/04/2026

Excited to be heading to Barcelona for #CHI2026 to host our workshop PoliSim: LLM Agent Simulation for Policy! This year, we’ve seen incredible interest from researchers across HCI, NLP, CSS, and Policy. We accepted 25 outstanding papers, with 5 selected as Best Paper nominees.

282

Nick Vincent @nickmvincent.bsky.social · 11/04/2026

So in practice, it's not a *complete* mystery, and a best case scenario would be a combination of retroactive dataset documentation + some attempt at retroactive attestation (but incomplete) + strong expectations for attestation going forward

000

Nick Vincent @nickmvincent.bsky.social · 11/04/2026

more detailed datasheets, and at least aiming for some kind of "coarse attestation" My read is that right now, people are assuming that pretraining for frontier models is pretty similar to open models like Olmo, just with more resources at each stage: filtering, fancy synthetic augmentation, etc.

100

Nick Vincent @nickmvincent.bsky.social · 11/04/2026

Oh yes definitely agree this will be a sticking point! My view is that we'll probably need to accept some kind of partial "declaration of bankruptcy" re: "first gen data". If attestation / data details become important signals, we could imagine AI developers at least releasing...

100

Nick Vincent @nickmvincent.bsky.social · 11/04/2026

New blog post: a more explicit vision of an “attestation-forward” data policy strategy for AI. With the right approach to attestations, I think we can get a win-win-win-win (for: consumers of AI products, auditors, AI developers worried about distillation, information quality):

210

Nick Vincent @nickmvincent.bsky.social · 06/04/2026

I think very funnily enough at the end of a long conference day my brain did a very similar merge on the interpretation side and parsed it as such (i.e. I assumed you were talking about the podcast with Josh 😂)

110

Nick Vincent @nickmvincent.bsky.social · 06/04/2026

Some thoughts on new policy paper from OpenAI: dataleverage.substack.com/p/people-fir... Lots of good stuff, it's great to foster more discussion of these kinds of ideas, and I think this particular set of ideas has the potential to really enable data flow as a powerful accountability lever!

030

Nick Vincent @nickmvincent.bsky.social · 28/03/2026

I totally mixed up the threading from phone, so went with the delete and repost strategy (but didn't repost yet 😂!). In fact, getting meta with the conference, I think I'll try to do a better organized recap using one of the longer form atproto tools and link to there!

120

Nick Vincent @nickmvincent.bsky.social · 27/03/2026

Thread for some misc thoughts / pins from @atproto.science talks and sessions: - big gap in demos that give potential users (eg scientists who used to post on other platforms) a “wow” moment for features that at proto enables (I had this recently with semble + margin interop)

120

Nick Vincent @nickmvincent.bsky.social · 27/03/2026

At UBC today for @atproto.science workshop! Much to discuss.

150

Reposted by Nick Vincent

ATProto Science @atproto.science · 26/03/2026

We have an exciting panel tomorrow @11am with @nickmvincent.bsky.social Laure Haak (@verime.coop) and Ellie DeSota (@metagov.bsky.social , SciOS)! The panel will explore how governance & sustainability challenges facing the broader atproto ecosystem are mirrored in its open science applications >

atmo.rsvp

Can decentralists cooperate? Rethinking commons and collective action in the age of platforms and AI

Event: Can decentralists cooperate? Rethinking commons and collective action in the age of platforms and AI

2104

Nick Vincent @nickmvincent.bsky.social · 23/03/2026

Also, a meta blog (very much inspired by teaching grad research communication this term and trying to answer very "dynamic questions") about the value of using coding agents to iterate on fancy sites vs. just sharing simple bullet points, plain text posts and spreadsheets

000

Nick Vincent @nickmvincent.bsky.social · 23/03/2026

Will likely have another blog with more thoughts on the emerging schemes for data licensing / preference signaling (and particular angle that I'm excited about, and think will actually happen: markets for attested evals and attested training data)

100

Nick Vincent @nickmvincent.bsky.social · 23/03/2026

I'm planning to post a blog explaining the purposes of the data counterfactuals site (aimed at highlighting connections between technically-focused and socially-driven data-centric work)

100

Nick Vincent @nickmvincent.bsky.social · 23/03/2026

Longer post(s) coming on these topics, but for those interested in "data counterfactuals" and various "data licenses" proposals I have made some big updates to two static site resources: datacounterfactuals.org and datalicenses.org

datacounterfactuals.org

Data Counterfactuals

An interactive explainer for a unifying frame across valuation, scaling, selection, poisoning, privacy, and collective action.

112

Nick Vincent @nickmvincent.bsky.social · 11/03/2026

I greatly enjoyed this conversation, and I think it might be interesting to a wide range of folks interested in AI, data, collective action, etc! Thanks for having me @tbsocialist.bsky.social (and structuring a really exciting conversation, and putting together such a nice final product!)

042

Reposted by Nick Vincent

Joshua Dávila @allmodelsarewrong.online · 11/03/2026

🚨Collective action strategies in the age of AI w/ Nick Vincent I spoke to @nickmvincent, AI researcher and author of the Data Leverage substack, about how AI systems are built on the collective output of humanity's digital labor and what we can do about it. FULL EPISODE⬇️

152

Reposted by Nick Vincent

Jason W. Burton @jasonburton.bsky.social · 11/03/2026

🧑‍💻 New paper at #chi2026 w @lorenzspreen.eurosky.social and @stefanherzog.bsky.social Are you worried about how social media algorithms affect people’s beliefs? We are, so we tested engagement-based ranking algorithms against alternatives in a pre-reg’d collaborative filtering experiment... 🧵

2199

Nick Vincent @nickmvincent.bsky.social · 11/03/2026

One q that’s come up — curious if you / the team have early thoughts on best way to handle bibliographic info. Eg say I want to make a static site with a “bib section”. I want to host the content “dynamically” in a Semble collection. Should I make cards that just link to semantic scholar or similar?

110

Nick Vincent @nickmvincent.bsky.social · 10/03/2026

Thoughts coming soon! So far: v excited about the interop and it was really an atproto-clicks-for-me moment (when my prev collection auto-populated, very exciting). Trying out github.com/treethought/... as my local editor so I can sync/unify collection and tag taxonomies across margin & semble

github.com

GitHub - treethought/obsidian-atmosphere: Surf the atmosphere from the coziness of your vault

Surf the atmosphere from the coziness of your vault - treethought/obsidian-atmosphere

010

Nick Vincent @nickmvincent.bsky.social · 09/03/2026

Two natural allies of a "Data Transparency" agenda: capabilities forecasters and social simulators --- dataleverage.substack.com/p/two-natura... (Short post reacting to recent discourse around AI forecasting + social simulation; both stand to independently benefit from "Clear Data Rules")

dataleverage.substack.com

010

Nick Vincent @nickmvincent.bsky.social · 05/03/2026

Pretty excited about trying move more of my info consumption through semble.so, margin.at, and www.graze.social Seems like there's some built in integration between @semble.so and margin -- very interested in figuring out a flow here (and further connect to local file workflows)

semble.so

Semble — A social knowledge network for research

Follow your peers' research trails. Surface and discover new connections. Built on ATProto so you own your data.

1123

Nick Vincent @nickmvincent.bsky.social · 17/02/2026

New post: thoughts on the paradox of reuse in 2026 after a few months of coding agent discourse. 'The Paradox of Reuse in 2026: A Case of Quasi-Enclosure, or "Subsidized Club Goods that Sort of Look Like Public Goods"' dataleverage.substack.com/p/the-parado...

dataleverage.substack.com

The Paradox of Reuse in 2026: A Case of Quasi-Enclosure, or "Subsidized Club Goods that Sort of Look Like Public Goods"

How we can understand, and react to, the complicated impacts of AI systems on online communities and knowledge commons

040

Reposted by Nick Vincent

Ronen Tamari @ronentk.me · 13/02/2026

1/ Finally wrote up “The Story of Mendeley”! Most people know the tool, few know about its rise and fall. The Mendeley story provides important clues for how to build self-sustaining AND non-extractive knowledge commons, which is why I think it deserves more attention 🧵

cairos.leaflet.pub

What happened to Science Goodreads and how do we rebuild it? A 65 million dollar question (at least) - CAIROS Blog

The story of the rise and fall of Mendeley

67437

Nick Vincent @nickmvincent.bsky.social · 27/01/2026

Yeah... completely agree on both sides 😅😬

010

Nick Vincent @nickmvincent.bsky.social · 27/01/2026

OpenAI launching an overleaf competitor seems like it could be a big deal, and particularly interesting in wake of the NeurIPS hallucinations discourse (an important issue, but a lot of the back-and-forth I saw seemed to be missing a lot of important factors): openai.com/index/introd...

openai.com

Introducing Prism

Accelerating science writing and collaboration with AI.

071

Nick Vincent @nickmvincent.bsky.social · 12/01/2026

Yes, def agree! I think local approach is becoming more viable, especially for personal use (but some tension if people are specifically using models just because they want to know what the capabilities frontier looks like)

030

Nick Vincent @nickmvincent.bsky.social · 11/01/2026

Writing a follow up post on data aspects of coding agents. One thing that's really under-discussed, IMO -- as far as I can tell, NO coding agent allows for consumer users to trigger server-side deletion of transcripts or even metadata. Anyone seen anything to the contrary?

361

Nick Vincent @nickmvincent.bsky.social · 09/01/2026

Seems plausible that some motivation for labs to restrict usage of subscription auth tokens is the value of structured data from using the official app, but unfortunate that the current data control for agents is super limited (30 days or 5 yrs, no indiv deletions, etc.)

010