Dan Simpson
@danpsimpson.bsky.social
I don’t know man, I just don’t know.
Pinned
Reposted by Dan Simpson
This is an excellent analogy, because my recollection from grade school is that the pen on the right looks fun and exciting, and then you play with it for a few minutes and realize it's not actually useful for anything and in fact makes some tasks more cumbersome, and never think about it again.
yeah, just out of interest, how many people choose the pen on the right for real work or art? See a lot of them in professional workplaces, do you?
January 22, 2026 at 12:21 PM
This is an excellent analogy, because my recollection from grade school is that the pen on the right looks fun and exciting, and then you play with it for a few minutes and realize it's not actually useful for anything and in fact makes some tasks more cumbersome, and never think about it again.
Reposted by Dan Simpson
Anania's experience with her trans identity
January 20, 2026 at 2:51 PM
Anania's experience with her trans identity
Reposted by Dan Simpson
(me making fun of your crop rotation idea and thereby holding our people back another 5000 years) jeff thinks the beans have to take turns lmao
August 7, 2023 at 6:38 AM
(me making fun of your crop rotation idea and thereby holding our people back another 5000 years) jeff thinks the beans have to take turns lmao
As an experiment, I spent this afternoon trying to do something mildly difficult and getting ChatGPT to help. It fucking sucked. That is all.
January 17, 2026 at 11:25 PM
As an experiment, I spent this afternoon trying to do something mildly difficult and getting ChatGPT to help. It fucking sucked. That is all.
Reposted by Dan Simpson
For audio, log-signature features complement Fourier-style features (e.g. spectrograms) in the sense that certain characteristics that are noisy or poorly captured by Fourier features are cleanly captured with certain log-signature features. Particularly interesting for audio source separation.
2/2
2/2
January 17, 2026 at 8:57 AM
For audio, log-signature features complement Fourier-style features (e.g. spectrograms) in the sense that certain characteristics that are noisy or poorly captured by Fourier features are cleanly captured with certain log-signature features. Particularly interesting for audio source separation.
2/2
2/2
Reposted by Dan Simpson
Recently, I have been exploring path log-signatures as features in machine learning models for audio. I found that no library supported backprop through log-signatures using modern pytorch, so I created a small library with minimal dependencies here: github.com/Froskekongen...
1/2
1/2
GitHub - Froskekongen/log_signatures_pytorch: Differentiable path signature and log-signature computation in pytorch
Differentiable path signature and log-signature computation in pytorch - Froskekongen/log_signatures_pytorch
github.com
January 17, 2026 at 8:54 AM
Recently, I have been exploring path log-signatures as features in machine learning models for audio. I found that no library supported backprop through log-signatures using modern pytorch, so I created a small library with minimal dependencies here: github.com/Froskekongen...
1/2
1/2
The problem with the Death Becomes Her recasting is that it should’ve been Bonnie Milligan.
January 17, 2026 at 2:32 AM
The problem with the Death Becomes Her recasting is that it should’ve been Bonnie Milligan.
Reposted by Dan Simpson
OK well they are. Hope that helps
FTC Chairman Andrew Ferguson says the agency is examining Big Tech's acqui-hires to make sure "they are not an attempt to get around" its merger review process (Bloomberg)
Main Link | Techmeme Permalink
Main Link | Techmeme Permalink
January 16, 2026 at 8:29 PM
OK well they are. Hope that helps
Anyway. My hot R vs Python take for 2026 is that I hate programming in Python because it desperately needs a more explicit type system. R suffers less from this.
January 16, 2026 at 1:55 PM
Anyway. My hot R vs Python take for 2026 is that I hate programming in Python because it desperately needs a more explicit type system. R suffers less from this.
In the bad old days of yore, data people would do their R vs Python rage-baiting. When really all it was was Everything vs pandas
the most remarkable thing to me about (python) pandas is how quickly i look for viable alternatives whenever i try to use it
January 16, 2026 at 1:45 PM
In the bad old days of yore, data people would do their R vs Python rage-baiting. When really all it was was Everything vs pandas
Reposted by Dan Simpson
I'm drafting a paper with a collaborator in which we are trying to sell a model with penalised splines. It's opened up a discussion on the 'correspondence' between penalties and priors.
Below are some excerpts from an email I sent to my coauthor. This is something I struggle to communicate.
Below are some excerpts from an email I sent to my coauthor. This is something I struggle to communicate.
January 14, 2026 at 2:38 PM
I'm drafting a paper with a collaborator in which we are trying to sell a model with penalised splines. It's opened up a discussion on the 'correspondence' between penalties and priors.
Below are some excerpts from an email I sent to my coauthor. This is something I struggle to communicate.
Below are some excerpts from an email I sent to my coauthor. This is something I struggle to communicate.
Reposted by Dan Simpson
TIL that Hal Sedgwick, Eve Sedgwick's husband, lived in Frank Rosenblatt's farmhouse attic at Cornell.
archive.ph/Q8EfZ
archive.ph/Q8EfZ
January 13, 2026 at 8:44 PM
TIL that Hal Sedgwick, Eve Sedgwick's husband, lived in Frank Rosenblatt's farmhouse attic at Cornell.
archive.ph/Q8EfZ
archive.ph/Q8EfZ
Reposted by Dan Simpson
FDA drugs and biologics guidance for Bayesian clinical trials is something I've been dreaming of for decades. I am thrilled that this draft has been released!! #Statistics #StatsSky #fda #clinicaltrial #bayes
This is a big deal for speeding up clinical trials, and for pediatric use after adult trials have proved successful. The @USFDA goes Bayesian.
A win for #IBD patients, and especially for kids with #IBD. @crohnsandcolitisau.bsky.social @f2harrell.bsky.social
www.fda.gov/news-events/...
A win for #IBD patients, and especially for kids with #IBD. @crohnsandcolitisau.bsky.social @f2harrell.bsky.social
www.fda.gov/news-events/...
FDA Issues Guidance on Modernizing Statistical Methods for Clinical Trials
The U.S. Food and Drug Administration today published draft guidance designed to facilitate the use of Bayesian methodologies in clinical trials of drugs and biologics, helping drug developers make be...
www.fda.gov
January 13, 2026 at 4:07 PM
FDA drugs and biologics guidance for Bayesian clinical trials is something I've been dreaming of for decades. I am thrilled that this draft has been released!! #Statistics #StatsSky #fda #clinicaltrial #bayes
When you hit one of Rust’s sharp edges it’s a _real experience_. (Do not ask me about sanitising error messages in binaries unless you have provided me with a case of bourbon)
January 13, 2026 at 7:06 PM
When you hit one of Rust’s sharp edges it’s a _real experience_. (Do not ask me about sanitising error messages in binaries unless you have provided me with a case of bourbon)
Oh my god the ability to get from the west to the east side without taking the 125th St bus ride of the damned!
Governor Hochul will announce an extension of the Second Avenue Subway to 125th and Broadway in her State of the State address on Tuesday. buff.ly/2qxRQs6
Westward Ho! Hochul Proposes to Extend Second Ave. Subway Along 125th Street to Broadway - Streetsblog New York City
The next step for the Second Avenue Subway is west across 125th Street to Broadway, Gov. Hochul will announce Tuesday.
nyc.streetsblog.org
January 13, 2026 at 7:03 PM
Oh my god the ability to get from the west to the east side without taking the 125th St bus ride of the damned!
Reposted by Dan Simpson
"without witness, without reward" is a really good principle to follow.
whether it's action, kindness, providing aid, or charity, you can do a lot of good in this world without seeking the validation that comes from the number on your post going up.
whether it's action, kindness, providing aid, or charity, you can do a lot of good in this world without seeking the validation that comes from the number on your post going up.
A lot of people on social media who insist you advertise every act of resistance you do aren't about actually resisting. They're about being SEEN resisting. And those are two very different things.
January 9, 2026 at 5:00 PM
"without witness, without reward" is a really good principle to follow.
whether it's action, kindness, providing aid, or charity, you can do a lot of good in this world without seeking the validation that comes from the number on your post going up.
whether it's action, kindness, providing aid, or charity, you can do a lot of good in this world without seeking the validation that comes from the number on your post going up.
New York is limited by only what you want to find. It’s a magical place if you’re open to it.
New York is diverse on a level that I have found it's hard to explain to someone who has lived in (e.g.) Europe their entire life and never traveled elsewhere.
I can cite statistics but they're so outside the realm of the familiar to people that they are too abstract to hold meaning.
I can cite statistics but they're so outside the realm of the familiar to people that they are too abstract to hold meaning.
January 9, 2026 at 7:16 AM
New York is limited by only what you want to find. It’s a magical place if you’re open to it.
Reposted by Dan Simpson
New York is diverse on a level that I have found it's hard to explain to someone who has lived in (e.g.) Europe their entire life and never traveled elsewhere.
I can cite statistics but they're so outside the realm of the familiar to people that they are too abstract to hold meaning.
I can cite statistics but they're so outside the realm of the familiar to people that they are too abstract to hold meaning.
January 9, 2026 at 2:30 AM
New York is diverse on a level that I have found it's hard to explain to someone who has lived in (e.g.) Europe their entire life and never traveled elsewhere.
I can cite statistics but they're so outside the realm of the familiar to people that they are too abstract to hold meaning.
I can cite statistics but they're so outside the realm of the familiar to people that they are too abstract to hold meaning.
I forgot to eat dinner so after this I went to a diner near my house which I did not realise was staffed exclusively by gorgeous bears at midnight on a Thursday.
January 9, 2026 at 5:56 AM
I forgot to eat dinner so after this I went to a diner near my house which I did not realise was staffed exclusively by gorgeous bears at midnight on a Thursday.
The most helpful assumptions are the explicitly expressed ones. Nuance is hard on here so I’ll just say it pithy: it’s exceptionally hard to do good, valid, and accurate statistical work without understanding and communicating the tradeoffs and assumptions you have made.
Which is why assumptions you can check aren't very helpful: if an assumption doesn't cost anything it won't buy much
January 8, 2026 at 6:34 PM
The most helpful assumptions are the explicitly expressed ones. Nuance is hard on here so I’ll just say it pithy: it’s exceptionally hard to do good, valid, and accurate statistical work without understanding and communicating the tradeoffs and assumptions you have made.
Reposted by Dan Simpson
Which is why assumptions you can check aren't very helpful: if an assumption doesn't cost anything it won't buy much
January 8, 2026 at 5:32 PM
Which is why assumptions you can check aren't very helpful: if an assumption doesn't cost anything it won't buy much
I think this framing would require more reflection than most ML luminaries show. This algorithms bias induced a huge reluctance to consider ways in which progress/problems in the field are data-driven. My hot take is that this is why people have such a hard time productionizing current methods
A history of ML would likely talk less about neural networks, MIT CSAIL and Minsky, McCarthy, etc., and talk more about the various ways in which datasets have played such a critical role in the development of pattern recognition methods.
January 7, 2026 at 4:00 PM
I think this framing would require more reflection than most ML luminaries show. This algorithms bias induced a huge reluctance to consider ways in which progress/problems in the field are data-driven. My hot take is that this is why people have such a hard time productionizing current methods
Mate. Let me tell you. NO ONE is more willing to rebrand/re-scope/absorb than CS researchers who have magically come across data (but not context for that data)
I remember, even within the introduction to AI course I audited in grad school in 2015, I remember the inclusion of some graph like this.
The thing which is maybe surprising in my reread is how quickly it seems like researchers within these field are willing to accept the redefinition in scoping.
The thing which is maybe surprising in my reread is how quickly it seems like researchers within these field are willing to accept the redefinition in scoping.
January 7, 2026 at 3:56 PM
Mate. Let me tell you. NO ONE is more willing to rebrand/re-scope/absorb than CS researchers who have magically come across data (but not context for that data)
Reposted by Dan Simpson
Once again asking the good people of Bluesky to understand that "generative art" is not the same thing as "AI art". The image below is generative art. There is no AI involved whatsoever. Adding generative artists to blocklists because you don't understand the distinction is a dick move
January 5, 2026 at 1:53 AM
Once again asking the good people of Bluesky to understand that "generative art" is not the same thing as "AI art". The image below is generative art. There is no AI involved whatsoever. Adding generative artists to blocklists because you don't understand the distinction is a dick move
Reposted by Dan Simpson
Night, all, and let us try to be good to one another — and ourselves.
January 6, 2026 at 6:31 AM
Night, all, and let us try to be good to one another — and ourselves.