Sign in

Mor Naaman

@informor.bsky.social
8.8K followers 1K following 995 posts

Cornell Tech professor (information science, AI-mediated Communication, trustworthiness of our information ecosystem). New York City. Taller in person. Opinions my own. mornaaman.com

PostsRepliesMedia
Mor Naaman @informor.bsky.social · 7h
Dang it here's a better link (without the ref to the paper section included in the previous one). www.nature.com/articles/s41... Also, my website now linked from my profile.
nature.com
Introducing AI to an online petition platform changed outputs but not outcomes - Nature Human Behaviour
Corpus et al. examine how the introduction of an AI writing tool impacted Change.org petitions, finding that it increased petition homogeneity but did not improve petition outcomes.
000
Mor Naaman @informor.bsky.social · 7h
Recent papers from @dgrand.bsky.social and others show that the persuasive power of AI comes from providing a high rate of facts/claims (true/not); this is unlikely to have diminished effect over time. On the other hand, AI language is likely to encounter more resistance IMO, but no evidence yet
000
Mor Naaman @informor.bsky.social · 7h
I think we did; @isabelcorpus.bsky.social can remind us... But as you point out, the potential general trend is exactly what we are controlling against (somewhat related: growth in participation resumed post Trump II).
020
Mor Naaman @informor.bsky.social · 9h
An inexcusable omission, now fixed! Thanks.
010
Mor Naaman @informor.bsky.social · 9h
Yes! Regression to the mean. Who returns for a second petition? Mostly those who had a successful first one.
110
Mor Naaman @informor.bsky.social · 9h
This too is an interesting hypothesis; from research that I have seen (and done) so far this is not yet the case. Maybe in the future...
120
Mor Naaman @informor.bsky.social · 9h
Yes very valid hypothesis. We don't *think* that this is what's happening here because the period we studied was not yet widespread in AI use or suspicion, but we definitely mention that as a possibility in the article. No good way (that we could think about) to find out.
100
Mor Naaman @informor.bsky.social · 10h
Paper and materials here; if you don't have access, you can find the unofficial pre-print version on my website. www.nature.com/articles/s41... Happy to hear feedback and ideas!
nature.com
Introducing AI to an online petition platform changed outputs but not outcomes - Nature Human Behaviour
Corpus et al. examine how the introduction of an AI writing tool impacted Change.org petitions, finding that it increased petition homogeneity but did not improve petition outcomes.
350
Mor Naaman @informor.bsky.social · 10h
Why was the AI not helpful? See the paper for more discussion, but some possible reasons (we are looking at now) include decreased specificity of petitions and (I think, more significantly) reduced effort by the petition writers, who were less committed after AI wrote for them.
130
Mor Naaman @informor.bsky.social · 10h
Isabel came up w/ another analysis, looking at repeat petition writers who returned to write a 2nd petition. We compared people who wrote 2 petitions before the AI launch w/ those who wrote 1 before and 1 after AI feature access, confirming our results: AI access changed the text, not the outcomes.
110
Mor Naaman @informor.bsky.social · 10h
One could ask whether these diminished outcomes are due to new authors joining the platform with the increased access that writing with AI enables (which is one of the reasons the feature was launched in the first place). We saw no evidence of a widespread increase in participation. Further...
110
Mor Naaman @informor.bsky.social · 10h
Was the AI assistance helpful? In general, petitions were NOT more successful with access to AI -- if anything, our results show that petitions may have been less successful in hitting signature and comment thresholds (see the end of this thread for speculation as to why this was the case).
121
Mor Naaman @informor.bsky.social · 10h
(In the supplementary materials, you can also see an increase, post-AI, in the use of words such as "implement", "urge", and "mandate" in the post-AI petitions' titles).
131
Mor Naaman @informor.bsky.social · 10h
As you can imagine, the petitions' textual/lexical aspects shifted with access to the AI feature: they were, on average, 50 words longer with access to "write with AI"; the sentences were longer, and the words were more complex. At the same time, the petitions became more similar to each other.
151
Mor Naaman @informor.bsky.social · 10h
If you are a data scientist/causal inference person, you know what that means: we had a natural experiment that could be leveraged using diff-in-diff analysis to estimate the causal impact of the AI feature on various aspects of the petitions.
150
Mor Naaman @informor.bsky.social · 10h
This was work led by the amazing @isabelcorpus.bsky.social with @allisonkoe.bsky.social and @eegilbert.org. Looking at the data we crawled (millions of petitions), Isabel noticed that the textual changes lagged in Australia; we later confirmed this was due to late launch of the AI feature there.
160
Mor Naaman @informor.bsky.social · 10h
🆕 📝 New paper officially out in @nathumbehav.nature.com! In 2023, Change.org built an AI-assisted writing tool into their platform. We offer causal evidence that access to the AI tool changed petition text (word count, language)... but may have actually hurt outcomes (signatures, comments).
13314
Mor Naaman @informor.bsky.social · 04/10/2026
* no pun intended (More on the course and my visits soon)
000
Mor Naaman @informor.bsky.social · 04/10/2026
While I reject the general idea of reanimated AI historical characters I also have to wholly endorse the outcome: nicely done video, and takeaways which align very well with the summary of my AI & Society course -- students in Bhutan and in Dhaka (so far) groked* it. www.nytimes.com/2026/09/30/o...
nytimes.com
Opinion | What Would John Maynard Keynes Say About A.I.? The Keynes Bot Told Us. (Gift Article)
Nearly a century ago, Keynes predicted that technology would liberate us from work. To imagine what he would think of artificial intelligence today, a Princeton economist used A.I.
340
Mor Naaman @informor.bsky.social · 17/09/2026
"many of the six incidents involved older A.I. models that were never deployed" according to OpenAI. That just puts your mind at ease, doesn't it. www.nytimes.com/2026/09/16/t...
nytimes.com
OpenAI Discloses Six New Incidents of ‘Concerning' A.I. Behavior (Gift Article)
The artificial intelligence company also released a framework for reporting when its systems go wrong.
100
Mor Naaman @informor.bsky.social · 11/09/2026
"Anger about AI has risen considerably. The share of respondents describing themselves as angry rose from 19% in December 2024 to 30% in August 2026. Worry remains the most common feeling Americans report about AI at 50%." naiom.net/reports/
naiom.net
Reports
REPORTS NAIOM Report #4: AI attitudes, work, and regulation in America 2026 KEY TAKEAWAYS Use of major AI tools among US adults rose from 50% in December 2024 to 72% in
020
Mor Naaman @informor.bsky.social · 11/09/2026
Yes, annoying. Does it not pick up context when you restart Claude in the same folder?
101
Mor Naaman @informor.bsky.social · 10/09/2026
Ok it's an important project and a great team of researchers, but what is the point of this story? It is one thing to report about papers or results before peer review. But this is just an ongoing work without any outcomes yet. I am so confused. Or maybe it's a warning to the models to play nice?
120
Mor Naaman @informor.bsky.social · 09/09/2026
*$22.5M more likely since it is probably using long context (more than 272K input tokens with previous agent conversation, though most of it cached). In any case, tens of millions. No wonder they forgo the Clay Mathematics Institute's $1M prize
000
Mor Naaman @informor.bsky.social · 09/09/2026
Math is a special use case, but this highlights the intense inequality that AI ushers in: rich companies (and countries/institutions) with access and enough resources to run the strongest models will see greater gains, leaving the rest of the world behind, doing (sometimes literally) manual work.
110
Mor Naaman @informor.bsky.social · 09/09/2026
One note about OpenAI's latest impressive math feat: according to them, it involved a total of 300 billion output tokens. While it's hard to estimate input load (not reported, sometimes cached), the output math is easier: these alone would cost $15M* under current newest model GPT-6 Astra.
231
Mor Naaman @informor.bsky.social · 02/09/2026
Another fun report from one of my favorite model-trackers, @simonwillison.net simonwillison.net/2026/Sep/2/c...
simonwillison.net
Claude’s new system prompt really doesn’t want to reproduce song lyrics
Anthropic publish the system prompts for their Claude consumer applications (Claude.ai and the Claude mobile apps—sadly not for Claude Cowork or Claude Code). I love that they do this, and …
040
Mor Naaman @informor.bsky.social · 02/09/2026
How much modification would you need to see in these parameters to cross the threshold? And, if these modifications are engineered/constructed by the model architects when they design the model, would that still count?
100
Reposted by Mor Naaman
Mor Naaman @informor.bsky.social · 28/08/2026
Great writeup, but unsatisfying just like the discussion of AI consciousness. Say we agree we need to see that "appraisal-related emotion-like representations produce sustained, multisystem reorganization of processing", who is to decide what is "sustained multisystem reorganization of processing"?
223
Mor Naaman @informor.bsky.social · 28/08/2026
The entire episode is a must-listen, and, while ostensibly about film, it is the deepest discussion of AI consciousness and rights you will find anywhere in the audio world. Give it a go. www.lawonfilm.com/blade-runner...
lawonfilm.com
Ep 61: Blade Runner & Blade Runner 2049 — Law on Film
022
Mor Naaman @informor.bsky.social · 28/08/2026
One of the highlight points in the episode: if we accept the idea of AI consciousness, and possibly then the idea of AI rights, we also accept the fact that the Tyrell Corporation can create infinite such replicants, aligned with the morals, attitudes, and well-being of its owners.
110
Mor Naaman @informor.bsky.social · 28/08/2026
In the meantime, other dangers lurk in our Replicant future. For a fantastic discussion of this topic, hear my brilliant colleague @frankpasquale.bsky.social discuss Blade Runner on the (surprisingly entertaining) podcast Law on Film by @jonathanhafetz.bsky.social www.lawonfilm.com/blade-runner...
lawonfilm.com
Ep 61: Blade Runner & Blade Runner 2049 — Law on Film
210
Mor Naaman @informor.bsky.social · 28/08/2026
These are untestable hypotheses that leave us with belief-based guidelines. We either believe AI/LLMs can have emotions/consciousness, or we do not. There is no evidence that can settle this, though certainly there is evidence that can shift people's attitudes on the matter. Maybe that's the point.
110
Mor Naaman @informor.bsky.social · 28/08/2026
Great writeup, but unsatisfying just like the discussion of AI consciousness. Say we agree we need to see that "appraisal-related emotion-like representations produce sustained, multisystem reorganization of processing", who is to decide what is "sustained multisystem reorganization of processing"?
223
Reposted by Mor Naaman
Justin Hendrix @justinhendrix.bsky.social · 24/08/2026
The 2027 Tech Policy Press Fellowship application is live! We’re seeking journalists, policy professionals, and researchers to pursue independent reporting and analysis on issues at the intersection of technology and democracy. Learn more and apply by September 24:
techpolicy.press
Call for Applications: 2027 Tech Policy Press Fellowship Program
2027 Tech Policy Press fellowship applications are due on September 24, 2026.
01217
Mor Naaman @informor.bsky.social · 24/08/2026
In 1984
0120
Reposted by Mor Naaman
Mor Naaman @informor.bsky.social · 22/08/2026
Excellent summary of the current state of knowledge -- unsurprising because Kai talked to some of the best in the business for this piece, authors and their critics.
1176
Mor Naaman @informor.bsky.social · 22/08/2026
Not missing but relevant is the power of models to shift your attitude in interactions beyond messaging or chat. As we showed in this work, even AI autocomplete suggestions can shift people's attitudes on important societal topics, often without their awareness: www.science.org/doi/10.1126/...
science.org
Biased AI writing assistants shift users’ attitudes on societal issues
Biased AI writing assistants shift people’s attitudes about societal issues; common interventions do not prevent this influence.
0119
Mor Naaman @informor.bsky.social · 22/08/2026
Excellent summary of the current state of knowledge -- unsurprising because Kai talked to some of the best in the business for this piece, authors and their critics.
1176
Mor Naaman @informor.bsky.social · 14/08/2026
For some reason I am reminded of this Shecky Greene joke Frank Sinatra once saved my life. A group of guys were beating me up. Frank suddenly showed up and said "ok guys, that's enough".
0120
Mor Naaman @informor.bsky.social · 14/08/2026
Sadly, horrific as it is, this event would not have gotten such exposure if there wasn't a US citizen involved. Also "the army’s struggle to curtail settler violence" is... misleading at best. The article details show the insane are running that asylum too (as the country, at least until Oct 27).
183
Mor Naaman @informor.bsky.social · 13/08/2026
Me too (August!) but will share!
020
Mor Naaman @informor.bsky.social · 13/08/2026
Polls open in Clacton where a ridiculously dressed novelty candidate clown is running against Count Binface
Farage. Photograph: Toby Shepheard/Reuters
061
Mor Naaman @informor.bsky.social · 12/08/2026
Which is the bigger point here, of course: we do not yet have the renewed capacity to send the critical signals we want to send in #AI-mediated-communication. Until we address this problem, these kinds of communications will be rendered close-to-useless "cheap talk".
110
Mor Naaman @informor.bsky.social · 12/08/2026
There is research to be done on when this self-disclosure may actually raise suspicion and cause a backfire (we know AI-use disclosures do). But in this domain, every such email I get is already suspect, so I believe the effect could be positive until they all start including this no-AI disclaimer.
100
Mor Naaman @informor.bsky.social · 12/08/2026
I get several PhD inquiries a day, but I truly appreciate this prospective PhD putting the key concern front and center. My website says "I do not guarantee a response, especially if your message doesn't clearly demonstrate that it wasn't authored by AI". So at least they are following directions.
Email subject including "(not authored by AI)"
161
Mor Naaman @informor.bsky.social · 10/08/2026
So disappointed in you. Followed you when you were punching up. Unfollowing now. Shame.
160
Mor Naaman @informor.bsky.social · 09/08/2026
Also handed out two Aces to his rival
060
Mor Naaman @informor.bsky.social · 09/08/2026
Look at all you *Bluesky* Cornellians!
020
Reposted by Mor Naaman
Kimmy W @ghostynewt.bsky.social · 07/08/2026
* 911 operator voice * I’m sorry to hear you’re having suicidal thoughts 😰 As a large language model, I’m not able to help. Instead, please consider calling a qualified 911 operator or speaking with a professional therapist. Hope your day gets better! Hang in there! 🤗” * hangs up on u *
241