Paul Stillman

@stillman

Behavioral Scientist studying self-control and motivation. Assistant Professor at Boston University

243 Following427 Followers151 Posts

YouTube Bluesky

Is this your profile?Claim it and stillman.dirty.social is yours.Claim profile

Paul Stillman@stillman· 13d

Some friends of mine wrote a paper that touched on this, and part of it was having people guess the prices of stuff from old sears catalogs. Massively underestimated how expensive everything used to be (and how much better things have gotten): psycnet.apa.org/record/2024-...

Abstract: In six studies, we find evidence of efficiency neglect: when thinking about the effects of population growth, people intuitively focus on increased demand while neglecting the changes in production efficiency that occur alongside, and often in response to, increased demand. In other words, people tend to think of others solely as consumers, rather than as consumers as well as producers. Efficiency neglect leads to beliefs that the real costs of some consumer goods are rising when they are actually decreasing and may contribute to antiimmigration sentiments because of the fear that increasing local population creates competition for fixed resources. We demonstrate that economic pessimism and antiimmigration sentiments are reduced when people are prompted to consider their own beliefs about increased productivity over time, but are unchanged when they consider their beliefs about increases in demand. Together, these findings shed light on people’s lay economic theories and suggest promising interventions. (PsycInfo Database Record (c) 2025 APA, all rights reserved)

Paul Stillman@stillman· 17d

Drake meme
Top panel: Measured use of LLMs for modest efficiency gains
Bottom panel: replace all humans with LLMs and severely degrade your product

Paul Stillman@stillman· 27d

Reminds me of this banger from the guy who spent 500 hours to get to level 99 in the first reactor of FF7

Life does not have inherent meaning; to say that our lives are pointless and our achievements meaningless is to state the obvious. No matter how grand our achievements or how broad their scope, time turns all to dust and death destroys all memory. But that does not mean we cannot ascribe our own meaning to what we do. It is because nothing has meaning unto itself that we are free to create meaning, to make metaphor, and in doing so reflect on ourselves and our world.

Leveling to 99 in the first reactor is pointless and meaningless. So why do I do it? I do it to express my hatred, and more importantly my disdain, for Dick Tree. I do it to express the camaraderie I feel for those of us who have followed this topic for years only to be disappointed by [Dick Tree]. I do it to prove to myself that I can persevere. The act is meaningless; I give it meaning.

Paul Stillman@stillman· Jun 23

Plot centers around finding and killing patient zero

Paul Stillman@stillman· Jun 15

Still from 12 Monkeys with Carol Florence and David Morse

Paul Stillman@stillman· Apr 24

Happened again with the next thing I asked it. This time the answer was right at least.

Claude with the prompt "Can I use separate_wider_delim (using _ as the separator) and specify that I want to drop the second column (it's always redundant)?" with the response:

Yes! Use the cols_remove argument — wait, that's not it. The right argument is too_many — no.
The correct answer is: separate_wider_delim has a cols argument to select which output columns to keep — no, it doesn't.
Let me be honest: the way to drop specific output columns in separate_wider_delim is via the names argument, where you can pass NA for any column you want to discard:
rdf |>
separate_wider_delim(x, delim = "_", names = c("alpha", NA))
Passing NA as a name for a position tells separate_wider_delim to drop that piece entirely. This is the cleanest way to do it within the call itself.

Paul Stillman@stillman· Apr 24

Has anyone else experienced Claude going in circles while trying to generate code? I feel like I've gotten variants on the following response quite a few times when trying to use it for relatively niche behavior in R (also the answer it gives here and lower in the post is wrong)

Chat with claude, where the prompt is "I am using unnest_wider on a column-list that contains an unnamed numeric vector, and I want to specify column names. How do I do that within the unnest_wider call?" and the response from claude is:

You can use the names_sep parameter combined with the names_repair approach, but the most direct way is the names_transform — actually, the right parameter is names_repair... let me be precise:
The correct parameter is names_repair — no, actually the one you want is:
Use the names_repair argument — no. The answer is ptype — no.
Let me be direct: use the names_to argument — actually the correct answer is:

✅ Use unnest_wider() with a named list, or use the names_sep + names_repair combo
For an unnamed numeric vector, the cleanest solution is the names_repair argument — but specifically, the dedicated way to supply names is:
rdf |>
unnest_wider(col, names_sep = "_")

Paul Stillman@stillman· Apr 4

Number go up

Pic showing the top users on bluesky by engagement, with Mrs. Betty Bowers in the #3 spot

Paul Stillman@stillman· Mar 15

Why go to theory town when you can go to

Storefront with sign that says “praxis city” (taken in Rotterdam).

Paul Stillman@stillman· Feb 23

comment of mercumeme saying 'Shoulda gotten the donkey tbh'

Paul Stillman@stillman· Jan 6

This quote is extremely backwards: everything he writes here is true, but these forces allow bad actors to avoid punishment-- they don't spur a high profile and expensive case to oust someone. The most sympathetic part of Gino's lawsuit was her going "you let this shit slide with other people"

Ackman saying "The problem with big institutions is that they are massive bureaucracies which occassionally put aside the core principles on which they are founded because they are embarrassed, because they are busy, and/or it is easier to ignore the harm they may cause to the little guy in the interest of protecting their reputation and/or avoiding further distraction to leadership. They also often assume that the little guy will eventually give up and/or run out of resources to keep fighting," he wrote. "I am going to make sure that doesn't happen here."

Paul Stillman@stillman· Jan 6

Looks like there's no new evidence, just the same tired 'scammers filled out my survey' (how the scammers knew to fill the survey out exactly so that they would confirm Gino's hypotheses has never been explained, probably because it's nonsense) + blaming research assistants. Comical.

On X, Ackman raised several critiques that Gino's lawyers have previously mounted. He argued that Gino is "A victim of some unintentional data errors made by some of the research associates that assisted her in her research" as well as "scammer(s)" who filled out surveys with corrupted data that showed up in some of her papers. He also argued that harvard failed to give gino adequate time and opportunity to defend herself. For example, he accused Alain Bonacossa, Harvard business school's research integrity officer, of telling gino that she could not hire her own forensic-research firm, and then denying in tryial that he had done so.

Paul Stillman@stillman· Nov 23

A movie that takes place where you’re from

Paul Stillman@stillman· Nov 22

I’m not sure how prominently twin studies feature in modern instruction but I read a ton about them as an undergrad/phd student. This suggests that a lot of these conclusions may need to be revisited

Paul Stillman@stillman· Nov 12

Using this approach, we find evidence for *both* impulse inhibition and dynamic competition. Notably, however, impulse inhibition occurs much less frequently (in only 26% of successful self-control decisions) and seems to be characteristic of impatient and present-focused individuals

Depiction of every larger-later decision in our dataset (N = 46774) colored by their hierarchical clustering assignment using a two-cluster partition. Trials identified as impulse inhibition (N = 12,309, 26.3%) are colored red, and trials identified as dynamic competition (N = 34,465, 73.7%) are colored blue. Bold lines represent the two cluster centers. Shaded rectangles in the corners correspond to the two response options, and all trajectories are recoded to terminate at the right response option. Figure to scale

Paul Stillman@stillman· Nov 12

Recent work suggests mouse movements can serve as a window into the real-time evolution of choice. Here, we develop a new analytical approach that integrates the spatial and temporal information, allowing us to test whether choices are more consistent with impulse inhibition vs. dynamic competition

Depiction of both the temporal evolution of choice (top panels) and how these would manifest as mouse trajectories (bottom panels) for both stage-based impulse inhibition (left panels) and integrative dynamic competition (right panels). (A) Stage-based impulse inhibition predicts an evolution of choice such that automatic processes are dominant early and drive participants towards temptations (initial arrows highlighted in red), before controlled processes come online and inhibit these impulses, directing participants back towards the goal (second arrows highlighted in grey). Here, we depict three possible evolutions in which the impulse varies in magnitude and timing of correction. (B) Integrative dynamic-competition models predict a noisy accumulation of evidence, such that once evidence has accumulated beyond a certain threshold (dashed lines), a choice is made. Such models do not predict temporal precedence to either goals or temptations. Here, we depict three possible evolutions that vary in their drift rate. (C) How the impulse inhibition depicted in (A) might manifest in mouse trajectories: movement towards the temptation proportional to the strength of the impulse, followed by a discrete correction back to the goal once controlled systems come online. (D) How dynamic competition depicted in (B) might manifest in mouse trajectories: graded trajectories in which the directness is proportional to the drift rate

Paul Stillman@stillman· Nov 12

Until now it has been difficult to test which of these accounts better describes self-control because modern tools do not capture the real-time dynamics of choice. To address this, we measured participants' computer mouse movements as they decided between immediate and delayed gratification

Three panels depicting a mouse-tracking experiment. The first panel shows a blank screen with a start button. After the participant clicks the start button, two options appear in the upper-left and upper right corners ($25 today and $40 in 180 days, respectively). The third panel depicts a participants' mouse trajectory, initially veering towards the smaller-sooner before deflecting back to choose the larger-later amount.

Paul Stillman@stillman· Nov 9