David Boetius
@davidboe.bsky.social
Reposted by David Boetius
Why not?
Blinding peer review, grant proposal review, etc. makes a lot of sense as to even the ground field given the difference pace around the globe.
I'm from 🇨🇴 which might rank in top for corruption. We may have improved as we call it clientelism or politics. Science is not immune to politics
Blinding peer review, grant proposal review, etc. makes a lot of sense as to even the ground field given the difference pace around the globe.
I'm from 🇨🇴 which might rank in top for corruption. We may have improved as we call it clientelism or politics. Science is not immune to politics
In light of the new OpenReview identity-leak scandal, it's a good time to question our assumptions about why blinding in peer review is helpful in the first place.
November 29, 2025 at 1:47 PM
Why not?
Blinding peer review, grant proposal review, etc. makes a lot of sense as to even the ground field given the difference pace around the globe.
I'm from 🇨🇴 which might rank in top for corruption. We may have improved as we call it clientelism or politics. Science is not immune to politics
Blinding peer review, grant proposal review, etc. makes a lot of sense as to even the ground field given the difference pace around the globe.
I'm from 🇨🇴 which might rank in top for corruption. We may have improved as we call it clientelism or politics. Science is not immune to politics
Reposted by David Boetius
Reposted by David Boetius
Handing back student work that’s been written by ChatGPT with a 0 followed by the comment “This essay will never stand in authentic wonder before the Beauty of God’s creation.”
Pope Leo XIV told students not to use artificial intelligence for homework, saying that AI ‘won’t stand in authentic wonder before the beauty of God’s creation.’
Even God Is Worried About ChatGPT
Pope Leo XIV told students not to use artificial intelligence for homework, saying that AI ‘won’t stand in authentic wonder before the beauty of God’s creation.’
www.vulture.com
November 26, 2025 at 3:35 PM
Handing back student work that’s been written by ChatGPT with a 0 followed by the comment “This essay will never stand in authentic wonder before the Beauty of God’s creation.”
Reposted by David Boetius
I believe eigen fly. I believe eigen touch the sky.
November 26, 2025 at 4:56 PM
I believe eigen fly. I believe eigen touch the sky.
Need to ensure stability of your neural ODE, but on a Riemannian manifold? 🤔 Worry no more! Our latest paper has you covered: robot-learning-control-lab.github.io/snmode-proje...
October 9, 2025 at 4:18 PM
Need to ensure stability of your neural ODE, but on a Riemannian manifold? 🤔 Worry no more! Our latest paper has you covered: robot-learning-control-lab.github.io/snmode-proje...
Cool paper on probabilistic verification using neuron splitting branch and bound! Uses Monte Carlo estimates of probabilities instead of computing the exact probabilities. IMO this is the main driver of the efficiency gain, as neuron splitting is too costly without it (previous works tried).
Fangji Wang, Panagiotis Tsiotras: BaB-prob: Branch and Bound with Preactivation Splitting for Probabilistic Verification of Neural Networks https://arxiv.org/abs/2509.25647 https://arxiv.org/pdf/2509.25647 https://arxiv.org/html/2509.25647
October 2, 2025 at 2:35 PM
Cool paper on probabilistic verification using neuron splitting branch and bound! Uses Monte Carlo estimates of probabilities instead of computing the exact probabilities. IMO this is the main driver of the efficiency gain, as neuron splitting is too costly without it (previous works tried).
Reposted by David Boetius
Juan Ramirez, Simon Lacoste-Julien: Dual Optimistic Ascent (PI Control) is the Augmented Lagrangian Method in Disguise https://arxiv.org/abs/2509.22500 https://arxiv.org/pdf/2509.22500 https://arxiv.org/html/2509.22500
September 29, 2025 at 6:35 AM
Juan Ramirez, Simon Lacoste-Julien: Dual Optimistic Ascent (PI Control) is the Augmented Lagrangian Method in Disguise https://arxiv.org/abs/2509.22500 https://arxiv.org/pdf/2509.22500 https://arxiv.org/html/2509.22500
Reposted by David Boetius
You may or my not have new messages
September 17, 2025 at 10:06 AM
You may or my not have new messages
Props to the person doing some actual work on this Saturday by mowing the overgrown lawn in front of my office while I sit at my desk
September 13, 2025 at 11:34 AM
Props to the person doing some actual work on this Saturday by mowing the overgrown lawn in front of my office while I sit at my desk
Good news! According to our weekly lab poll the internet wasn’t dead last week. 📈 But with only two votes. 2️⃣
September 13, 2025 at 9:30 AM
Good news! According to our weekly lab poll the internet wasn’t dead last week. 📈 But with only two votes. 2️⃣
Reposted by David Boetius
Just submitted the manuscript for the 2nd edition of Swarm Robotics: A Formal Approach 🎉
✅ ~2× the size
✅ fully updated with latest research
✅ ready4the breakthrough years ahead
Thanks to all who supported! Now in production—stay tuned.
Now it's in production – stay tuned.
#swarm #robotics #book
✅ ~2× the size
✅ fully updated with latest research
✅ ready4the breakthrough years ahead
Thanks to all who supported! Now in production—stay tuned.
Now it's in production – stay tuned.
#swarm #robotics #book
September 9, 2025 at 7:57 AM
When you go to the court and hit an unexpected three 🏀🏀🏀 then turn around and notice you‘re alone… This is how LRMs feel when they write to their reasoning buffer
August 26, 2025 at 6:24 PM
When you go to the court and hit an unexpected three 🏀🏀🏀 then turn around and notice you‘re alone… This is how LRMs feel when they write to their reasoning buffer
Interesting paper!
Liv Gorton, Owen Lewis: Adversarial Examples Are Not Bugs, They Are Superposition https://arxiv.org/abs/2508.17456 https://arxiv.org/pdf/2508.17456 https://arxiv.org/html/2508.17456
August 26, 2025 at 9:26 AM
Interesting paper!
Reposted by David Boetius
Contemplating this post will make you enlightened about LLMs in software
Interesting LLM translation failure (or success?) mode: Grok auto-translates a rot13 cipher that was being used to hide spoilers
August 25, 2025 at 12:11 AM
Contemplating this post will make you enlightened about LLMs in software
Reposted by David Boetius
This doesn't seem like an accurate summary to me. They tested CGP and LLM-assisted ES separately without combining the results (although they mention doing that in the future). And the improved performance is just compared to the Kalman filter, in scenarios where the Kalman filter is non-optimal
August 20, 2025 at 2:05 AM
This doesn't seem like an accurate summary to me. They tested CGP and LLM-assisted ES separately without combining the results (although they mention doing that in the future). And the improved performance is just compared to the Kalman filter, in scenarios where the Kalman filter is non-optimal
Reposted by David Boetius
Harvard researchers developed a framework combining Cartesian Genetic Programming and Large Language Models to discover new Kalman Filter variants. This method yields algorithms that excel beyond traditional ones, a major advance in scientific computing. https://arxiv.org/abs/2508.11703
Data-Driven Discovery of Interpretable Kalman Filter Variants through Large Language Models and Genetic Programming
ArXiv link for Data-Driven Discovery of Interpretable Kalman Filter Variants through Large Language Models and Genetic Programming
arxiv.org
August 19, 2025 at 11:30 PM
Harvard researchers developed a framework combining Cartesian Genetic Programming and Large Language Models to discover new Kalman Filter variants. This method yields algorithms that excel beyond traditional ones, a major advance in scientific computing. https://arxiv.org/abs/2508.11703
Nice colormap I didn’t get to use thus far 🍭 #UseBatlow
Scales (such as a colour bar) that are meant to represent one dimension of a data set can’t be used to highlight data ranges.
Instead, use superposed contour lines instead if you must know the individual values clearly on a map.
Here’s a crude example of how this map could be shown.
📊 #UseBatlow
Instead, use superposed contour lines instead if you must know the individual values clearly on a map.
Here’s a crude example of how this map could be shown.
📊 #UseBatlow
August 19, 2025 at 12:22 AM
Nice colormap I didn’t get to use thus far 🍭 #UseBatlow
Reposted by David Boetius
Scales (such as a colour bar) that are meant to represent one dimension of a data set can’t be used to highlight data ranges.
Instead, use superposed contour lines instead if you must know the individual values clearly on a map.
Here’s a crude example of how this map could be shown.
📊 #UseBatlow
Instead, use superposed contour lines instead if you must know the individual values clearly on a map.
Here’s a crude example of how this map could be shown.
📊 #UseBatlow
October 8, 2024 at 7:16 AM
Scales (such as a colour bar) that are meant to represent one dimension of a data set can’t be used to highlight data ranges.
Instead, use superposed contour lines instead if you must know the individual values clearly on a map.
Here’s a crude example of how this map could be shown.
📊 #UseBatlow
Instead, use superposed contour lines instead if you must know the individual values clearly on a map.
Here’s a crude example of how this map could be shown.
📊 #UseBatlow
Reposted by David Boetius
Well … the Hierarchical Reasoning Model didn’t quite hold up under scrutiny, at least not its central claim, I.e., its brain inspired architecture: arcprize.org/blog/hrm-ana...
But … from now on, HRM is going to be my best teaching material for why AI models should be open.
But … from now on, HRM is going to be my best teaching material for why AI models should be open.
August 16, 2025 at 12:23 PM
Well … the Hierarchical Reasoning Model didn’t quite hold up under scrutiny, at least not its central claim, I.e., its brain inspired architecture: arcprize.org/blog/hrm-ana...
But … from now on, HRM is going to be my best teaching material for why AI models should be open.
But … from now on, HRM is going to be my best teaching material for why AI models should be open.
Reposted by David Boetius
I can’t get this post out of my head
August 15, 2025 at 12:39 AM
I can’t get this post out of my head
Missing o4-mini-high :(
August 14, 2025 at 10:47 PM
Missing o4-mini-high :(
Inductive bias… I know where the inductive comes from but why is it called „bias“? 🤔 What’s the relation to biased estimators?
August 10, 2025 at 6:21 PM
Inductive bias… I know where the inductive comes from but why is it called „bias“? 🤔 What’s the relation to biased estimators?
Something cool to know about matrix multiplication: even matrix mult behaves unexpectedly in 100+ dimensions arxiv.org/abs/1412.6572
Explaining and Harnessing Adversarial Examples
Several machine learning models, including neural networks, consistently misclassify adversarial examples---inputs formed by applying small but intentionally worst-case perturbations to examples from ...
arxiv.org
August 10, 2025 at 7:43 AM
Something cool to know about matrix multiplication: even matrix mult behaves unexpectedly in 100+ dimensions arxiv.org/abs/1412.6572
Reposted by David Boetius
*chomsky voice* this isn’t even my normal form
August 9, 2025 at 5:50 PM
*chomsky voice* this isn’t even my normal form