Hokin
@hokin.bsky.social
Philosopher, Scientist, Engineer
https://hokindeng.github.io/
https://hokindeng.github.io/
November 7, 2025 at 5:38 AM
‼️ Video models start to reason, let's build-in-public scaled eval together 🚀
github.com/hokindeng/VM... (Apache 2.0) offers
1⃣One-click inference across ALL available models
2⃣Unified API & datasets & auto resume + error handling + eval
3⃣Plug new models and tasks in <5 lines of code
a thread (1/n)
github.com/hokindeng/VM... (Apache 2.0) offers
1⃣One-click inference across ALL available models
2⃣Unified API & datasets & auto resume + error handling + eval
3⃣Plug new models and tasks in <5 lines of code
a thread (1/n)
November 4, 2025 at 9:34 PM
‼️ Video models start to reason, let's build-in-public scaled eval together 🚀
github.com/hokindeng/VM... (Apache 2.0) offers
1⃣One-click inference across ALL available models
2⃣Unified API & datasets & auto resume + error handling + eval
3⃣Plug new models and tasks in <5 lines of code
a thread (1/n)
github.com/hokindeng/VM... (Apache 2.0) offers
1⃣One-click inference across ALL available models
2⃣Unified API & datasets & auto resume + error handling + eval
3⃣Plug new models and tasks in <5 lines of code
a thread (1/n)
Excited to share my essay with @carrot0817.bsky.social, Kaia Gao on the representational substrate of world-reasoning in both humans and machines has been accepted to the SpaVLE Workshop at #NeurIPS2025✨
a thread (1/n)
a thread (1/n)
November 2, 2025 at 11:50 PM
Excited to share my essay with @carrot0817.bsky.social, Kaia Gao on the representational substrate of world-reasoning in both humans and machines has been accepted to the SpaVLE Workshop at #NeurIPS2025✨
a thread (1/n)
a thread (1/n)
Very cool work !
"Intuitions of mathematical curves in young children's drawings"
In this new paper from the lab, Lorenzo Ciccione, Marie Lubineau, Theo Morfoisse and I show that 5 and 6 year olds already possess intuitions of linearity, curvature, period and compositionality.
www.sciencedirect.com/science/arti...
In this new paper from the lab, Lorenzo Ciccione, Marie Lubineau, Theo Morfoisse and I show that 5 and 6 year olds already possess intuitions of linearity, curvature, period and compositionality.
www.sciencedirect.com/science/arti...
November 1, 2025 at 4:10 PM
Very cool work !
best Halloween costume this year
Gonna wear this to the next faculty meeting to see if they think I’m Steve Pinker.
November 1, 2025 at 2:33 AM
best Halloween costume this year
Reposted by Hokin
Josh Tenenbaum's inspiring keynote at #cogsci2025 on growing vs scaling AI, the big questions of cognitive science, and the many open questions for the field.
August 1, 2025 at 11:14 PM
Josh Tenenbaum's inspiring keynote at #cogsci2025 on growing vs scaling AI, the big questions of cognitive science, and the many open questions for the field.
Reposted by Hokin
this was a hilarious final slide that just hung up there during q&a
Susan Carey mic-drop at #cogsci2025. "There are no innate concepts: Discuss"
August 2, 2025 at 3:39 PM
this was a hilarious final slide that just hung up there during q&a
Reposted by Hokin
Susan Carey mic-drop at #cogsci2025. "There are no innate concepts: Discuss"
August 2, 2025 at 3:22 PM
Susan Carey mic-drop at #cogsci2025. "There are no innate concepts: Discuss"
#CoreCognition #LLM #multimodal #GrowAI We spent 3 years to curate 1503 classic experiments spanning 12 core concepts in human cognitive development and evaluated on 230 MLLMs with 11 different prompts for 5 times to get over 3.8 millions inference data points.
A thread (1/n) - #ICML2025 ✅
A thread (1/n) - #ICML2025 ✅
June 30, 2025 at 6:07 AM
#CoreCognition #LLM #multimodal #GrowAI We spent 3 years to curate 1503 classic experiments spanning 12 core concepts in human cognitive development and evaluated on 230 MLLMs with 11 different prompts for 5 times to get over 3.8 millions inference data points.
A thread (1/n) - #ICML2025 ✅
A thread (1/n) - #ICML2025 ✅
What is the best testbed for human-like cognition in machines?🤔 Hardly anything comes better than gaze understanding.
Humans are extremely good at gaze reading 👁️ , and this ability develops extremely early in childhood.
Here, we tested 111 VLMs and 65 humans ⬇️
Humans are extremely good at gaze reading 👁️ , and this ability develops extremely early in childhood.
Here, we tested 111 VLMs and 65 humans ⬇️
👁️ 𝐂𝐚𝐧 𝐕𝐢𝐬𝐢𝐨𝐧 𝐋𝐚𝐧𝐠𝐮𝐚𝐠𝐞 𝐌𝐨𝐝𝐞𝐥𝐬 (𝐕𝐋𝐌𝐬) 𝐈𝐧𝐟𝐞𝐫 𝐇𝐮𝐦𝐚𝐧 𝐆𝐚𝐳𝐞 𝐃𝐢𝐫𝐞𝐜𝐭𝐢𝐨𝐧?
Knowing where someone looks is key to a Theory of Mind. We test 111 VLMs and 65 humans to compare their inferences.
Project page: grow-ai-like-a-child.github.io/gaze/
🧵1/11
Knowing where someone looks is key to a Theory of Mind. We test 111 VLMs and 65 humans to compare their inferences.
Project page: grow-ai-like-a-child.github.io/gaze/
🧵1/11
June 13, 2025 at 1:58 PM
What is the best testbed for human-like cognition in machines?🤔 Hardly anything comes better than gaze understanding.
Humans are extremely good at gaze reading 👁️ , and this ability develops extremely early in childhood.
Here, we tested 111 VLMs and 65 humans ⬇️
Humans are extremely good at gaze reading 👁️ , and this ability develops extremely early in childhood.
Here, we tested 111 VLMs and 65 humans ⬇️
Reposted by Hokin
Beautiful to see this initiative from a group of like minded PhD students collaborating together! 🚀
New Paper Alert ‼️ Current VLMs completely fail human gaze understanding 🙀 and scaling does NO help ‼️
However, humans, since an extremely age 🧒, are extremely sensitive to other people's gaze 🙄 👀
No mentors, no labs, only pre-doc students, 111 VLMs, and we did it 😎
However, humans, since an extremely age 🧒, are extremely sensitive to other people's gaze 🙄 👀
No mentors, no labs, only pre-doc students, 111 VLMs, and we did it 😎
June 11, 2025 at 11:49 PM
Beautiful to see this initiative from a group of like minded PhD students collaborating together! 🚀
New Paper Alert ‼️ Current VLMs completely fail human gaze understanding 🙀 and scaling does NO help ‼️
However, humans, since an extremely age 🧒, are extremely sensitive to other people's gaze 🙄 👀
No mentors, no labs, only pre-doc students, 111 VLMs, and we did it 😎
However, humans, since an extremely age 🧒, are extremely sensitive to other people's gaze 🙄 👀
No mentors, no labs, only pre-doc students, 111 VLMs, and we did it 😎
June 11, 2025 at 11:21 PM
New Paper Alert ‼️ Current VLMs completely fail human gaze understanding 🙀 and scaling does NO help ‼️
However, humans, since an extremely age 🧒, are extremely sensitive to other people's gaze 🙄 👀
No mentors, no labs, only pre-doc students, 111 VLMs, and we did it 😎
However, humans, since an extremely age 🧒, are extremely sensitive to other people's gaze 🙄 👀
No mentors, no labs, only pre-doc students, 111 VLMs, and we did it 😎
I think there is a fundamental division between “physical reasoning” and “intuitive physical reasoning” in human mind.
The former is grounded by System 2 where the latter is actually System 1.
The physics/graphics I have never thought of - and never thought about this in context of aphantasia
The former is grounded by System 2 where the latter is actually System 1.
The physics/graphics I have never thought of - and never thought about this in context of aphantasia
Out now in TiCS, something i've been thinking about a lot:
"Physics vs. graphics as an organizing dichotomy in cognition"
(by Balaban & me)
relevant for many people, related to imagination, intuitive physics, mental simulation, aphantasia, and more
authors.elsevier.com/a/1lBaC4sIRv...
"Physics vs. graphics as an organizing dichotomy in cognition"
(by Balaban & me)
relevant for many people, related to imagination, intuitive physics, mental simulation, aphantasia, and more
authors.elsevier.com/a/1lBaC4sIRv...
June 3, 2025 at 1:10 AM
I think there is a fundamental division between “physical reasoning” and “intuitive physical reasoning” in human mind.
The former is grounded by System 2 where the latter is actually System 1.
The physics/graphics I have never thought of - and never thought about this in context of aphantasia
The former is grounded by System 2 where the latter is actually System 1.
The physics/graphics I have never thought of - and never thought about this in context of aphantasia
Reposted by Hokin
It's an interesting phenomenon that some of the deepest questions about how life works have become what looks like impossibly obscure molecular biology stuck right at the back of Nature, which will never get covered by the science media. Like this. /1
www.nature.com/articles/s41...
www.nature.com/articles/s41...
DNA-guided transcription factor interactions extend human gene regulatory code - Nature
A large-scale analysis of DNA-bound transcription factors (TFs) shows how the presence of DNA markedly affects the landscape of TF interactions, and identifies composite motifs that are recognized by ...
www.nature.com
May 31, 2025 at 6:06 PM
It's an interesting phenomenon that some of the deepest questions about how life works have become what looks like impossibly obscure molecular biology stuck right at the back of Nature, which will never get covered by the science media. Like this. /1
www.nature.com/articles/s41...
www.nature.com/articles/s41...
Doing research not in your previous areas comes at a cost. Good paper but I disagree on the corollary.
Instead, the best scientists should always walk out their comfort zones. And indeed they always do.
www.nature.com/articles/s41...
Instead, the best scientists should always walk out their comfort zones. And indeed they always do.
www.nature.com/articles/s41...
The pivot penalty in research - Nature
An analysis of millions of scientific papers and patents reveals a ‘pivot penalty’ when researchers shift direction, with the impact of studies decreasing rapidly the further they move from their prev...
www.nature.com
May 29, 2025 at 3:21 PM
Doing research not in your previous areas comes at a cost. Good paper but I disagree on the corollary.
Instead, the best scientists should always walk out their comfort zones. And indeed they always do.
www.nature.com/articles/s41...
Instead, the best scientists should always walk out their comfort zones. And indeed they always do.
www.nature.com/articles/s41...
Reposted by Hokin
Check out our new work on making robots process touch more like brains!
Surprisingly, ConvRNNs best matched mouse cortex—passing the NeuroAI Turing Test. We also developed tactile-specific SSL augmentations and an Encoder-Attender-Decoder framework unifying ConvRNNs, SSMs & Transformers.
Surprisingly, ConvRNNs best matched mouse cortex—passing the NeuroAI Turing Test. We also developed tactile-specific SSL augmentations and an Encoder-Attender-Decoder framework unifying ConvRNNs, SSMs & Transformers.
1/ What if we make robots that process touch the way our brains do?
We found that Convolutional Recurrent Neural Networks (ConvRNNs) pass the NeuroAI Turing Test in currently available mouse somatosensory cortex data.
New paper by @Yuchen @Nathan @anayebi.bsky.social and me!
We found that Convolutional Recurrent Neural Networks (ConvRNNs) pass the NeuroAI Turing Test in currently available mouse somatosensory cortex data.
New paper by @Yuchen @Nathan @anayebi.bsky.social and me!
May 28, 2025 at 12:31 AM
Check out our new work on making robots process touch more like brains!
Surprisingly, ConvRNNs best matched mouse cortex—passing the NeuroAI Turing Test. We also developed tactile-specific SSL augmentations and an Encoder-Attender-Decoder framework unifying ConvRNNs, SSMs & Transformers.
Surprisingly, ConvRNNs best matched mouse cortex—passing the NeuroAI Turing Test. We also developed tactile-specific SSL augmentations and an Encoder-Attender-Decoder framework unifying ConvRNNs, SSMs & Transformers.
This is such a beautiful and important work. I have been thinking about the increasing automation due to AI a lot.
Aran here puts into actions and give us such a beautiful work!
It’s definitely part of the duties us as scientists to think about these important societal issues.
Aran here puts into actions and give us such a beautiful work!
It’s definitely part of the duties us as scientists to think about these important societal issues.
Can a Universal Basic Income (UBI) become feasible—even if AI fully automates existing jobs and creates no new ones?
We derive a closed-form UBI threshold tied to AI capabilities that suggests it's potentially achievable by mid-century even under moderate AI growth assumptions:
We derive a closed-form UBI threshold tied to AI capabilities that suggests it's potentially achievable by mid-century even under moderate AI growth assumptions:
May 27, 2025 at 9:18 PM
This is such a beautiful and important work. I have been thinking about the increasing automation due to AI a lot.
Aran here puts into actions and give us such a beautiful work!
It’s definitely part of the duties us as scientists to think about these important societal issues.
Aran here puts into actions and give us such a beautiful work!
It’s definitely part of the duties us as scientists to think about these important societal issues.
Sam is 100% correct on this. Indeed, human babies have essential cognitive priors such as permanence, continuity, and boundary of objects, 3D Euclidean understanding of space, etc.
We spent 2 years to systematically to examine and show the lack of such in MLLMs: arxiv.org/abs/2410.10855
We spent 2 years to systematically to examine and show the lack of such in MLLMs: arxiv.org/abs/2410.10855
May 24, 2025 at 5:55 AM
Sam is 100% correct on this. Indeed, human babies have essential cognitive priors such as permanence, continuity, and boundary of objects, 3D Euclidean understanding of space, etc.
We spent 2 years to systematically to examine and show the lack of such in MLLMs: arxiv.org/abs/2410.10855
We spent 2 years to systematically to examine and show the lack of such in MLLMs: arxiv.org/abs/2410.10855
Reposted by Hokin
Sharing a new paper from the lab. This paper, led by Sangyoon Ko, represents a merging of two longstanding research themes in the lab-- adult neurogenesis and systems consolidation.
rdcu.be/el18q
A short thread follows for those interested.
1/n
rdcu.be/el18q
A short thread follows for those interested.
1/n
Systems consolidation reorganizes hippocampal engram circuitry
Nature - A study shows that loss of memory precision associated with systems consolidation can be explained by neurogenesis-dependent reorganization of engram circuitry within the hippocampus over...
rdcu.be
May 14, 2025 at 4:32 PM
Sharing a new paper from the lab. This paper, led by Sangyoon Ko, represents a merging of two longstanding research themes in the lab-- adult neurogenesis and systems consolidation.
rdcu.be/el18q
A short thread follows for those interested.
1/n
rdcu.be/el18q
A short thread follows for those interested.
1/n
Reposted by Hokin
Hippocampal reactivation of planned trajectories is required for effective goal choice in an allocentric memory task https://www.biorxiv.org/content/10.1101/2025.05.10.653115v1
May 12, 2025 at 12:15 PM
Hippocampal reactivation of planned trajectories is required for effective goal choice in an allocentric memory task https://www.biorxiv.org/content/10.1101/2025.05.10.653115v1
Reposted by Hokin
If I had more free time, I would become a sci-fi novelist. I have a long list of plots that I want to write. Here's one.
May 10, 2025 at 9:29 AM
If I had more free time, I would become a sci-fi novelist. I have a long list of plots that I want to write. Here's one.
Reposted by Hokin
Lovely work — and yet another proof that perceptual integration, likely based on recurrent connections, can occur without awareness during the attentional blink.
See also www.pnas.org/doi/full/10....
See also www.pnas.org/doi/full/10....
May 6, 2025 at 6:13 AM
Lovely work — and yet another proof that perceptual integration, likely based on recurrent connections, can occur without awareness during the attentional blink.
See also www.pnas.org/doi/full/10....
See also www.pnas.org/doi/full/10....
Reposted by Hokin
Loving the new ICLR blogpost track. Here's a cool one on engrams in neuroscience vs mechanistic interpretability: d2jud02ci9yv69.cloudfront.net/2025-04-28-e...
In Search of the Engram in LLMs: A Neuroscience Perspective on the Memory Functions in AI Models | ICLR Blogposts 2025
Large Language Models (LLMs) are enhancing our daily lives but also pose risks like spreading misinformation and violating privacy, highlighting the importance of understanding how they process and st...
d2jud02ci9yv69.cloudfront.net
May 5, 2025 at 8:06 PM
Loving the new ICLR blogpost track. Here's a cool one on engrams in neuroscience vs mechanistic interpretability: d2jud02ci9yv69.cloudfront.net/2025-04-28-e...
Finally finish making our website 💐 growing-ai-like-a-child.github.io 🤩
with @carrot0817.bsky.social @marstin.bsky.social @tonyfeng.bsky.social @zoryzhang.bsky.social and many others @growai.bsky.social🚶♂️🚶♀️🚶
with @carrot0817.bsky.social @marstin.bsky.social @tonyfeng.bsky.social @zoryzhang.bsky.social and many others @growai.bsky.social🚶♂️🚶♀️🚶
GrowAI
Growing AI like a Child, at Scale Humans never "learn" intelligence. Humans develop intelligence. Biological lives on this planet take heavy advantage of intelligent primitives embedded in their genes...
growing-ai-like-a-child.github.io
May 4, 2025 at 11:51 PM
Finally finish making our website 💐 growing-ai-like-a-child.github.io 🤩
with @carrot0817.bsky.social @marstin.bsky.social @tonyfeng.bsky.social @zoryzhang.bsky.social and many others @growai.bsky.social🚶♂️🚶♀️🚶
with @carrot0817.bsky.social @marstin.bsky.social @tonyfeng.bsky.social @zoryzhang.bsky.social and many others @growai.bsky.social🚶♂️🚶♀️🚶
