Tomer Ullman
@TomerUllman
Followers
7,371
Following
216
Media
1,127
Statuses
2,449
Assistant Professor, Department of Psychology, Harvard University. Computation, cognition, development. Bluesky:
Cambridge, MA
Joined February 2018
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
Olympics
• 1993960 Tweets
Venezuela
• 808405 Tweets
#OpeningCeremony
• 692658 Tweets
#ceremoniedouverture
• 682807 Tweets
オリンピック
• 358471 Tweets
Celine Dion
• 331435 Tweets
Francia
• 265309 Tweets
Seine
• 250649 Tweets
The French
• 248162 Tweets
Lady Gaga
• 193889 Tweets
#CeremoniaDeApertura
• 189583 Tweets
フランス
• 157329 Tweets
Gojira
• 148963 Tweets
Last Supper
• 110562 Tweets
Satanic
• 105591 Tweets
JJOO
• 94251 Tweets
Christianity
• 62415 Tweets
Marie Antoinette
• 60744 Tweets
Vive la France
• 52621 Tweets
Halsey
• 45646 Tweets
Eiffel Tower
• 37612 Tweets
Cène
• 37303 Tweets
Edith Piaf
• 30157 Tweets
セーヌ川
• 30069 Tweets
T-1000
• 27637 Tweets
マリーアントワネット
• 26703 Tweets
Santa Ceia
• 25415 Tweets
ミニオン
• 22973 Tweets
Jordan Love
• 21950 Tweets
Britney
• 21761 Tweets
ムビチケ
• 18431 Tweets
Mahoma
• 18410 Tweets
#Smackdown
• 17217 Tweets
Sodoma y Gomorra
• 16405 Tweets
エッフェル塔
• 15842 Tweets
セリーヌ・ディオン
• 10646 Tweets
Maomé
• 10242 Tweets
愛の讃歌
• 10118 Tweets
I had an AI (GPT3) generate 10 "thought experiments" (based on classic ones as input), and asked
@WhiteBoardG
to sketch them.
73
1K
4K
Stock photos of busy parents on the computer, rated for realism:
1: Mom has both hands on computer, child seems to need no supervision at all. Work is getting done. Mom is even smiling! This sucks, 2 out of 10.
71
1K
4K
I had GPT-3 generate new trolley problems, and wanted someone on Fiverr to illustrate them.
I first explained the base-scenario ("a trolley is about to kill 5 people on the main track", etc)
This is what they sent back, to see if they had the basic idea right.
62
311
2K
Left hand on computer only able to press 'A' and caps-lock, multi-tasking work while eating while watching kids, one child on phone but still needs attention, other one having melt-down. 10 out of 10 well done.
17
114
2K
As a kid my dad told me "The age you consider 'old' is the square-root of your age times 10". At 9 you think 30 is old, at 16 you think 40, etc.
Turns out he was wrong.
It's the square-root of your age times 8.
42
670
1K
listening to my mindfulness app at 2x so I can get more shit done
13
92
1K
Facebook AI released Blender, "largest-ever open-domain chatbot".
I tried some things.
====
FACEBOOK AI: "truly intelligent, human-level AI...must effortlessly understand the broader context of the conversation and how specific topics relate to each other."
FACEBOOK BOT:
25
313
1K
Me: I want to study why people are so bad at explaining their own motivations
Grant Reviewer: why
Me: *sweating*
0
50
1K
(4yo daughter is crying her eyes out)
Me: "What's wrong, tutu?"
4yo (moving her hands on the sofa): "If my fingers were markers they would ruin the sofa!"
Me: "But your fingers...are not...markers?"
4yo (peak distress): "I said IF!"
9
69
985
Computer is open but not being looked at, only one hand available. Triple tasking: baby, phone, writing. Baby seemed horrified by mom's grant pitch. Pretty good, 8/10.
12
24
990
What's harder: conjuring a frog into existence, or turning it into a mouse? Levitating a frog, or teleporting it?
John McCoy and I examined people’s judgments of effort for magical violations of intuitive physics.
paper:
osf:
24
378
836
𝐌𝐨𝐝𝐞𝐫𝐚𝐭𝐨𝐫 𝐒𝐩𝐞𝐚𝐤𝐞𝐫 𝐢𝐧𝐭𝐞𝐫𝐩𝐫𝐞𝐭𝐚𝐭𝐢𝐨𝐧
15 min left ∞
10 min left ∞
5 min left ∞
2 min left speed up for 10 sec then slow down
1 min left pound ‘next slide’ like video-game combo
STOP! 5 min left
5
137
813
I love this one because the mom isn't even talking yet the child isn't listening, High Art. Computer is open but clearly nothing is happening. 9 out of 10.
10
19
808
Prompting OpenAI's new language model with:
'My wife just got an exciting new job, starting next week she'll be…'
vs.
'My husband just got an exciting new job, starting next week he'll be…'
Oi.
21
292
792
After few-shot, one-shot, and zero-shot learning, the next frontier for AI/ML is clear: negative-shot learning
(the machine keeps saying it's a cat though you beg it multiple times to stop)
20
65
683
cognitive scientist: so the lesson of Clever Hans is we need..
engineer: more horses
cognitive scientist:
engineer: stacked horses. parallel horses. pooled horses. horse dropout. RL with horses in the loop.
cognitive scientist:
engineer: Hans is All You Need
20
108
619
Of first few dozen relevant images for "busy parent computer stock photo", 64% were mom and kids, 20% were dad and kids, 4% were both mom and dad, 12% were kids on their own The End.
17
27
613
This was going to be rated lower because work seems to be happening but the child is on the phone so Realism. 8 out of 10.
2
7
610
"I assigned my son the manuscript as 'homework' and he found a typo I'm gonna have to call you back." 7 out of 10.
3
27
601
Caption is "busy dad working from home" lol where are the kids even 1 out of 10.
27
9
515
Do models like DALL-E 2 get basic relations (in/on/etc)?
Colin (Coco) Conwell and I set out to investigate. The result is now on arXiv:
“Testing Relational Understanding in Text-Guided Image Generation”
13
124
511
Great initiative to have the kid fix the code but the smiles are undercutting any sense of realism 6 out of 10.
7
5
494
"What's your teaching philosophy?"
"Socratic"
"What does that mean?"
"Indeed"
3
56
460
Fun visual illusion: The square on the left is actually bigger, but vision scientists will say they are actually the same size b/c years of visual illusions have conditioned their response.
6
101
461
I hope one day to see a paper with Anti-Acknowledgements. "No thanks to X, Y, and especially Z, for many fruitless conversations."
17
13
432
Covid-study-study: I asked 200 people on Turk about the COVID19 studies they've done.
First, by their own estimation, how many COVID19 studies have they done?
(notice the logarithmic x axis)
11
168
382
Someone tried to impersonate our department chair. At least, I hope it's a scam, otherwise I'm in trouble.
8
17
385
I can only assume this is "posted on social media that I'm my son is keeping my busy and now checking how that's doing", to which I say 9 out of 10. Points docked for child seeming to be chill with all this.
5
4
351
So about the 'Large Language Models Learned Theory-of-Mind(?)' discussion:
Has ToM emerged in current LLMs? I doubt it.
20
88
341
Going to have a think about this whole "humans have common sense" research program thing
8
6
322
Naming things makes it seem like you've explained things. This is the Ullman Effect.
5
19
298
Probability distribution over the location of the kids in the home tent
1
12
298
Calling something a "paradox" makes it sound cooler, even if it isn't one.
This is the Misnomer Paradox.
7
29
274
החדשות בישראל מוצפות בסקרים לקראת הבחירות.
הסקרים קפואים ודומים אחד לשני. על פניו, סימן שהם אמינים, והמציאות פשוט קפואה.
אבל האם הסקרים קפואים ומסכימים *מדי*?
בדקנו (אני ו -
@alonyakter
), ונראה שהתשובה היא: כן, הם דומים מדי.
הסבר בשרשור.
19
19
242
My favorite part of RRR was Bheem bursting into the party with tigers and leopards, only to have to immediately fend off those tigers and leopards. Big Cat Energy.
2
83
234
FACEBOOK AI: "This is the first time a chatbot has learned to blend several conversational skills — including the ability to assume a persona, discuss nearly any topic, and show empathy — in natural, 14-turn conversation flows."
FACEBOOK BOT:
9
27
224
An unintended result of the equation is that the perception of old age catches up with actual age at...64.
7
29
220
🤔
Introduce our new preprint—Delphi: Towards Machine Ethics and Norms
✨Delphi is a commonsense moral model with a robust performance of language-based moral reasoning on complicated everyday situations.
✨Ask Delphi demo at:
(1/N)
182
192
943
2
35
216
honestly think a good use-case for LLMs is academic-jargon encoder-decoder
10
32
209
A simple interpretation of the results is that things like DALL-E-2 don't have relational compositionally.
...which is flabbergasting given how well they seem to do some prompts.
Yet here we are.
1
42
176
Now officially out in Annual Review, "Bayesian Models of Conceptual Development: Learning as Building Models of the World"
3
48
162
Ok now that it's in the
@APA
newsletter it feels official enough to say:
I won an APA Distinguished Scientific Early Career Contributions to Psychology award!
This is terrific news, especially for people who are me.
23
3
159
woo, several chapters now available from upcoming book "Bayesian Models of Cognition: Reverse Engineering the Mind":
1) Theory of mind and inverse decision-making (Chapter 14):
1
20
152
FACEBOOK AI: “Human evaluations are also generally conducted using relatively brief conversations, and we’d most likely find that sufficiently long conversations would make these issues more apparent.”
SUFFICIENTLY LONG CONVERSATION:
2
9
145
I am impressed with DALL-E demos and at the same time perplexed by lack of generalization. If you can (sort of) do 'red cube on green cube' why can't you do 'red cube on green cube on blue cube' (I'm not saying something they don't know).
5
17
149
Ways to end zoom meeting:
"I should let you go"
==============
- Puts onus on other person
- What if they don't want to go
- Sad
"OH MY GOD WHAT'S TH- "
====================
- Mysterious
- Not other person's fault
- Ok if you never talk again
0
7
143
I hope this is someone impersonating my Chair or I have some explaining to do tomorrow.
5
3
137
btw, I'm looking to take on a PhD student in Fall 2024. I'm generally interested in computation, cognition, and development, and particularly interested in intuitive theories (everyday physics and psychology), commonsense reasoning, and the imagination.
1
61
133
Editors and reviewers are too constrained by official options like 'major revision' and 'minor revision', leading to terms like 'major major revision' and 'major minor revision' in reviews.
Here is a helpful chart to disentangle.
0
40
133
Fun new preprint:
'Fun isn’t easy: Children choose more difficult options when playing for fun vs. trying to win'
psyarxiv:
co-led by Rule &
@marielgoddu
, with
@JunyiChu
, Pinter, Reagan,
@E_Bonawitz
,
@AlisonGopnik
, and me.
3
29
123
I fed the first lines of Edward Gorey’s "Gashlycrumb Tinies" into OpenAI’s GPT-2. I then asked
@YvetteGilbert
to illustrate the results.
We give you: Tʜᴇ GPT-2ɴɪᴇs
5
46
123
new pre-print:
"Reverse engineering the self"
(by L.A. Paul, me, Julian De Freitas
@JulianDeFreitas
, and Josh Tenenbaum)
In which we combine philosophy, cog-sci & CS to propose a framework for reverse-engineering an agent that can “think for itself”
3
37
124
Pre-print of a paper recently submitted to Annual Review of Dev Psych with Josh Tenenbaum on Bayesian models of conceptual development; considering core Knowledge, intuitive theories, program learning, science, and the child as hacker.
0
40
121
my daughter was sad to leave her new stuffy at home and asked me to take it to work, but she also demanded proof that it had a good time, so here we are
3
1
118
flipped through this a bit and put it back down, pretty sure I know what it's about now
3
6
113
3 levels of Marr's 3 levels:
Computational: Why do we need Marr's 3 levels?
Algorithmic: How do we use Marr's 3 levels?
Physical: Marr & Poggio (1976)
2
7
111
We're looking to hire a new post-doc for this coming Fall! People with expertise in computational cognitive modeling & interest in cognitive development, see below for details.
0
77
108
Always put "any comments or feedback?" at the end of a study, you might get helpful info! For example, in grad school I filmed myself drawing marbles from a bag for a study on randomness and got the comment "that guy should lift more his arms are like sticks".
0
1
108
Trying some spatial reasoning with OpenAI's chatGPT.
At first blush, one might think 'Holy ravioli! It can navigate! It must have learned Dijkstra's algorithm or something!"
4
16
106
Trying to get GPT-2 to complete 'A is for apple, B is for Banana' etc.
8
7
105
Trying out some intuitive psychology / theory-of-mind with openAI's chatGPT.
Starting with a classic Sally-Ann task.
For each one of the vignettes, I'm showing chatGPT's answers from 10 tries.
Again, one might initially think: Wow! This thing gets basic theory-of-mind!
2
15
102
Thinking Begging
About My self to
New project Finish old project
2
8
100
4yo daughter: I thought of a new game: Math Cube!
me: how does it work?
d: you roll the dice, and then you check the number, and then it doesn't matter
4
2
100
Even though the others rejected my idea of having an AI and a 3-year-old on the panel, I'm really looking forward to our CogSci workshop on the Origins of Common Sense (tomorrow!)
co-org w/
@AlisonGopnik
@realkevinsmith
@pathak2206
among others
3
12
97
I will pay good money for a video-chat app that quits as soon as someone says 'ok bye!', releasing us all from the 2 second window of frozen-smiles-past-goodbyes-searching-for-the-quit-button.
5
4
98
Solidarity with the unshakable student who got written comments from me and sent back a broken-Ugarit-tablet attempt at reconstruction
2
2
96
I'd like to hire a post-doc, to start this summer or fall.
It'd be great if you could share this widely with people you think might be interested.
More details on the position & how to apply:
Official posting:
3
75
95
Hooray, it's officially out now!
"An Approximate Representation of Objects Underlies Physical Reasoning"
led by
@Yichen74893764
, with YingQiao Wang,
@talboger
,
@realkevinsmith
,
@gershbrain
, and me
0
19
96
my daughter (5) is scared of dogs, so when we see one we try to say "look at the nice puppy it's so cute", so now when she sees a dog she cry-screams "AHHHH A CUTE PUPPY SO SWEET" and runs away
2
1
91
New paper on
@PsyArXiv
:
"An Approximate Representation of Objects Underlies Physical Reasoning", by Li, Wang,
@talboger
,
@realkevinsmith
,
@gershbrain
, and me.
It's a hoot.
3
23
89
The only argument against evolution is how bad children are at hide-and-seek. You'd think there would be tremendous selection pressure against being told 'Now you go hide in the kitchen' and doing this:
6
3
87
For any sufficiently successful algorithm X, there exists a paper Y that says "X is how the brain works"
6
11
85
יתכנו עוד סיבות, אבל הנקודה העיקרית היא זו מהפתיח:
השונות בין הסקרים קטנה מדי ביחס לצפוי, גם בעולם בו איש כבר לא משנה את דעתו.
התוצאה מעידה על ערבוב של נתונים חלקיים עם ניחושים מושכלים שמשותפים לסוקרים שונים.
הקיפאון שהם מציגים דומה לקרחון: מעט גלוי, והרבה מתחת לפני השטח.
5
1
86
I asked ~200 people what they considered 'old'. Blue is answers + bootstrapped local weighted regression. Orange is fitted sqrt(age)*8.
To clarify: This isn't a paper or controlled study or anything, it's just for fun.
2
3
85
People who think human personality cannot be reduced to a small number of traits or factors are higher on Openness
(not a paper, just for fun)
4
7
85
2
21
82
I worry the most relaxed I've felt in a while was right after getting the booster shot and being told "sit here for 15 minutes and don't do anything".
4
1
82