Really interesting article! I am especially intrigued by your identification of the tooling gap as an obstacle to progress. I firmly believe that researchers need a new primitive object type at a higher level of abstraction than what currently exists. That would support all of the endeavors you describe here.
Every table, every figure, every research output should contain the metadata allowing it to be traced back to the raw data it originated from.
Only by utilizing this metadata - containing the structure of data processing pipelines - will storytelling tools be able to handle the variety and specificity of workflows necessary for research grade analytics.
For example, in one of my recent Python projects I created a Thunk class (https://github.com/ResearchOS/ross_thunk), inspired by the Thunk type in Haskell. In both Haskell and my Python implementation, Thunks are simply wrappers around functions and their inputs and outputs. They can be called and otherwise interacted with exactly the same as the values they wrap. However, crucially, they also contain references to the Thunk objects that come before them and after them in the data processing pipeline, preserving the Directed Acylic Graph structure of the data processing pipeline for other tools to use, while minimally modifying researchers’ existing analytics workflows.
I’m very curious to hear your thoughts on whether something like this Thunk implementation is a concrete step in the right direction towards building tools to bridge the tooling gap between computations and storytelling.
When I was in grad school, there was a forum of PhD students across labs in the university that we used to communicate and share work. Would love to see this digitally and across schools.
Is there a world in which you see some kind of forum for researchers only? Or even a world where outside observers could see conversations taking place (like this one) without being littered by bots and slop?
This is great. I also had a Slack channel from uni where people would share their work, but most engagement ended up being pretty dry/dead and didn't translate to actual marketing in the sense of engaging with the real world outside the channel's bubble.
I think this is what early Twitter and Bluesky tried to be. There's also something to be said about the type of researcher who engages with others; usually PIs, older, less likely to adopt a new platform, or they communicate behind closed doors via email/institution specific platforms.
The question is what would qualify someone to "participate" in such a group, and your inclusion/exclusion criteria might be taut. Is someone who's published a pre-print qualified? Does a MS student qualify? Would this create a strange split in public/research (which we're trying to prevent)?
Lovely essay. I have tons of thoughts, but let me stop with two:
1. What might research look like if a team/collaboration started with the press release as Amazon teams were famous for doing? Would it change the course of that research or only help market it better? Would we choose research problems differently if the final output was critiqued before you formulated the first hypothesis? Of course, research is meant to be original and unpredictable, but how might we assess its predicted value differently from the way it's done in grant applications.
2. We might be returning to the beginnings of science, where private patrons were crucial to the success of early modern scientists. Telling a compelling story was crucial to success. Galileo demoed his telescope (and the moons of Jupiter) to the Pope. Even though that got him into trouble later, it was partly because of narrative power; his 'Dialogue Concerning the Two Chief World Systems' was a great piece of literature.
1) One of the things I did want to include was a 'changelog' that many application-layer teams are known to do. This also signals back to what the oldest journals did; document the changes to research over time, and encourage collaboration because it was an "open call" to the challenges that researchers were facing at the time and maybe someone with a specific skillset would read it, and help out. I've heard suggestions of revising journals this way; of encouraging teams to "show their work" unfold (also to fight back against plagiarism and replicated results). What stopped me was like what you mentioned, but also thoughts on 1) would teams be ready to share this type of information, with fierce competition and IP concerns (especially if housed at a university) floating, and 2) would they see this as useful/would this be different from regular paper publications? Is there a new audience that would enjoy this who isn't already reading the papers?
There’s precedent: DARPA program managers already write “Heilmeier Catechism”-style briefs before funding — a kind of proto-press release. NIH and NSF grant abstracts function like this too, but often end up in bureaucratic language rather than compelling narrative. So maybe it’s less about forcing a prediction of results, and more about forcing clarity on why this problem matters and to whom.
2) I agree with this strongly. You see this with philanthropy now, where, although it was much more distributed than say the Medici family, researchers will flock to them for capital and freedom to experiment. Similarly, Robert Boyle’s “literary technology” in the New Experiments used vivid prose and diagrams to convince readers of the air pump’s legitimacy (Steven Shapin and Simon Schaffer write about this in Leviathan and the Air-Pump).
Would love to hear more of your thoughts - this was loads of fun to write.
On 1) Funders have to lead the way by demanding transparent formats, not just for data collection but also for hypothesis formation and reasoning. We are making LLMs do this, why not humans :) Of course it also needs new forms of recognition for those intermediate steps. Especially for research on domains such as climate change where we can't afford to be slow.
On 2) perhaps we need to explicitly bring back the tradition of Natural Philosophy - there's a reason why Hobbes' Leviathan and Boyles' research on the Air Pump were read by the same people. The larger point being: we can't expect disruptive forms of documentation and reasoning to arise without new cognitive styles that disrupt existing disciplinary boundaries. At the very least you want researchers who can experiment with new hypotheses, new documentation patterns and new social incentives at once. That feels very Natural Philosophy with a helping of Design to me.
Do you think we’re starting to see a new cohort of researchers who are willing to do this today — experimenting not just with hypotheses, but also with documentation styles and incentive systems in public? A few names come to mind for me on the margins of 'thinking' (ex. the Michael Nielsens and Nadia Asparouhovas of the world), but I wonder who stands out for you as having done this well, and more on the lab/applied side.
Bret Victor comes to mind, though less of a researcher and more of a designer. I see murmurings here and there, but it needs explicit coordination to become a thing IMHO, and deliberate cultivation of these skill sets by a community of researchers. Might be worth putting together a workshop!
My hunch (and hope) is that a lot of the most promising ideas here will start as integrations inside existing products/tools — extensions to arXiv, overlays on PDFs, plugins for Benchling/Notion/Slack — so that researchers can try them in situ, give early feedback, and we can see what actually sticks. Then, if something clearly works, it can evolve into a standalone layer.
I feel like we need a new job role in academia to just do all of this and manage it at department/school/faculty etc level. So it becomes as easy as updating a spreadsheet for the researcher/PI. Build a KPI around it etc. Then it is covered minimum level for all and those that want to excel and lean into this area can. Not all scientists will be comms stars and that's OK imo.
Some labs I’ve visited have a designated graphic designer that goes into this, which is fine when the teams are small, but then leads to backlogs and makes the smallest edits/changes a huge back and forth. Ultimately this becomes unsustainable IMO - increases friction, than if we just had simple plug and play tools that let everyone (researcher) control the narrative, look, tone, feel the way they like it.
It’s a bit like ghostwriting for tweets and using AI and stuff. Generally with writing, each person is uniquely different and it’s the little nuances that would benefit from having individual control over.
This all has to start with a shift in the scientific culture. Right now, there are only two recognized currencies for academic scientists:
1. Grants funded
2. Peer-reviewed papers
So, there seems to be no good incentive to communicate with the public. Time spent in public engagement is time not spent writing grant applications, doing experiments, or writing papers — the very products that one currently needs for career advancement.
We need to start training grad students skills in public engagement, and ensure that universities also value it (i.e., in tenure and promotion decisions). The scientific ecosystem needs PIs to produce original research, but it also needs people who can effectively communicate it.
I don't know the answer to this. I mostly work in the world of teaching clinical students (doctor of physical therapy), and have not been all that connected to typical PhD / post-doc training environments in recent years.
So, I think a typical grad program has room for some mentorship with communication. For example, the process is not done at the completion of a peer-review paper, but also extends to the next step of public engagment.
Conversely, I think most clinical programs (like DPT, MD, DVM, etc.) don't have as much flexibility to do this, even though it would be valuable. They are geared at creating entry-level clinicians, which involves preparing for a national-level licensing exam. So, the curriculum may be a bit more "rigid." When communication is emphasized, it's often based on communicating with individual-level patients, rather than broad public engagement.
One of the courses I teach touches on this, and includes an assignment in which they need to create a 3-5 minute answering a clinical question, aimed at a general audience. So we focus on select aspects, like ensuring the language isn't too technical (unnnecessary medical terminology, p-values, etc.), and trying to remain concise, yet accurate in their summary. It's not a comprehensive course on public engagment, but it's a start.
So, long answer, I don't know what grad programs are doing this well. Interested to hear others' responses.
I am not sure if I am on track where you mentioned of who is "experimenting not just with hypotheses, but also with documentation styles and incentive systems in public?" but Arcadia Science https://www.arcadiascience.com/ comes to mind in a not for profit but also non-academic mix pioneering new tooling, research and publication styles. I can't say there is a clear success or not, but an interesting experiment of how to create, perform, and communicate research.
> That convergence of tech (a vertical) and mainstream culture (horizontal) didn’t just happen because the code got better. It happened because the tooling to make it legible got better. Mac OS, GitHub, Figma, Twitter — each of them lowered the lift of distribution, making technical work culturally relevant — enough to catch the eye of even Anna Wintour
This is bullshit word salad. WTF does it even mean? macOS has been around for 40 years and I doubt the average Vogue employee or reader has any idea what GitHub or Figma is. Bezos caught Anna Wintour’s eye for the same reason Elon did: they’re the two richest men on the planet, not because she took a fancy for AWS.
Think you're missing the point here. The point I'm making is that there are 3 waves of tooling and distribution. Bezos and Elon caught attention because they're rich, but how did they get rich? 1) building a useful tool (the code, which I address), but also 2) knowing how to sell. And selling means knowing your user's pain points, marketing it to them well (or hiring a team that can).
For this to even have gotten to the attention of a Vogue employee means that tech did something right. For it to have crossed domains like this - outside of the world of code and software - was about legibility.
I don't think so. Amazon Studios is a major player in the Hollywood ecosystem so of course Vogue is going to suck up to him; plus Bezos loves hanging out with celebrities (see: launching Katy Perry into space).
Bill Gates, Larry Page, Sergey Brin, and Steve Jobs were never invited to the Met Gala, even though their products are more ubiquitous than Amazon and are likely used by every single Vogue employee every single day. And Jobs was 100x the salesman Bezos ever will be. Amazon minus AWS is not a tech company but Walmart on the web.
The Met Gala is only one example that emphasizes that tech has been so deeply embedded into culture/pop culture. I could reference Zuckerberg and having a film produced about him (in a somewhat negative light, so doubt he had a studio or financial influence over this), Apple having advertisements and endorsements with A-list celebrities, however medium you want to argue, the impressive thing is they’ve broken through in a way research hasn’t.
What are tools you would want to see exist?
Really interesting article! I am especially intrigued by your identification of the tooling gap as an obstacle to progress. I firmly believe that researchers need a new primitive object type at a higher level of abstraction than what currently exists. That would support all of the endeavors you describe here.
Every table, every figure, every research output should contain the metadata allowing it to be traced back to the raw data it originated from.
Only by utilizing this metadata - containing the structure of data processing pipelines - will storytelling tools be able to handle the variety and specificity of workflows necessary for research grade analytics.
For example, in one of my recent Python projects I created a Thunk class (https://github.com/ResearchOS/ross_thunk), inspired by the Thunk type in Haskell. In both Haskell and my Python implementation, Thunks are simply wrappers around functions and their inputs and outputs. They can be called and otherwise interacted with exactly the same as the values they wrap. However, crucially, they also contain references to the Thunk objects that come before them and after them in the data processing pipeline, preserving the Directed Acylic Graph structure of the data processing pipeline for other tools to use, while minimally modifying researchers’ existing analytics workflows.
I’m very curious to hear your thoughts on whether something like this Thunk implementation is a concrete step in the right direction towards building tools to bridge the tooling gap between computations and storytelling.
When I was in grad school, there was a forum of PhD students across labs in the university that we used to communicate and share work. Would love to see this digitally and across schools.
Is there a world in which you see some kind of forum for researchers only? Or even a world where outside observers could see conversations taking place (like this one) without being littered by bots and slop?
although, I will also add, the migration of academics off Twitter was largely due to noise and slop, so there is a good point there.
This is great. I also had a Slack channel from uni where people would share their work, but most engagement ended up being pretty dry/dead and didn't translate to actual marketing in the sense of engaging with the real world outside the channel's bubble.
I think this is what early Twitter and Bluesky tried to be. There's also something to be said about the type of researcher who engages with others; usually PIs, older, less likely to adopt a new platform, or they communicate behind closed doors via email/institution specific platforms.
The question is what would qualify someone to "participate" in such a group, and your inclusion/exclusion criteria might be taut. Is someone who's published a pre-print qualified? Does a MS student qualify? Would this create a strange split in public/research (which we're trying to prevent)?
Lovely essay. I have tons of thoughts, but let me stop with two:
1. What might research look like if a team/collaboration started with the press release as Amazon teams were famous for doing? Would it change the course of that research or only help market it better? Would we choose research problems differently if the final output was critiqued before you formulated the first hypothesis? Of course, research is meant to be original and unpredictable, but how might we assess its predicted value differently from the way it's done in grant applications.
2. We might be returning to the beginnings of science, where private patrons were crucial to the success of early modern scientists. Telling a compelling story was crucial to success. Galileo demoed his telescope (and the moons of Jupiter) to the Pope. Even though that got him into trouble later, it was partly because of narrative power; his 'Dialogue Concerning the Two Chief World Systems' was a great piece of literature.
Love this comment.
1) One of the things I did want to include was a 'changelog' that many application-layer teams are known to do. This also signals back to what the oldest journals did; document the changes to research over time, and encourage collaboration because it was an "open call" to the challenges that researchers were facing at the time and maybe someone with a specific skillset would read it, and help out. I've heard suggestions of revising journals this way; of encouraging teams to "show their work" unfold (also to fight back against plagiarism and replicated results). What stopped me was like what you mentioned, but also thoughts on 1) would teams be ready to share this type of information, with fierce competition and IP concerns (especially if housed at a university) floating, and 2) would they see this as useful/would this be different from regular paper publications? Is there a new audience that would enjoy this who isn't already reading the papers?
There’s precedent: DARPA program managers already write “Heilmeier Catechism”-style briefs before funding — a kind of proto-press release. NIH and NSF grant abstracts function like this too, but often end up in bureaucratic language rather than compelling narrative. So maybe it’s less about forcing a prediction of results, and more about forcing clarity on why this problem matters and to whom.
2) I agree with this strongly. You see this with philanthropy now, where, although it was much more distributed than say the Medici family, researchers will flock to them for capital and freedom to experiment. Similarly, Robert Boyle’s “literary technology” in the New Experiments used vivid prose and diagrams to convince readers of the air pump’s legitimacy (Steven Shapin and Simon Schaffer write about this in Leviathan and the Air-Pump).
Would love to hear more of your thoughts - this was loads of fun to write.
At what point do you think the DARPA/NIH bits were in their prime and what went wrong?
On 1) Funders have to lead the way by demanding transparent formats, not just for data collection but also for hypothesis formation and reasoning. We are making LLMs do this, why not humans :) Of course it also needs new forms of recognition for those intermediate steps. Especially for research on domains such as climate change where we can't afford to be slow.
On 2) perhaps we need to explicitly bring back the tradition of Natural Philosophy - there's a reason why Hobbes' Leviathan and Boyles' research on the Air Pump were read by the same people. The larger point being: we can't expect disruptive forms of documentation and reasoning to arise without new cognitive styles that disrupt existing disciplinary boundaries. At the very least you want researchers who can experiment with new hypotheses, new documentation patterns and new social incentives at once. That feels very Natural Philosophy with a helping of Design to me.
Do you think we’re starting to see a new cohort of researchers who are willing to do this today — experimenting not just with hypotheses, but also with documentation styles and incentive systems in public? A few names come to mind for me on the margins of 'thinking' (ex. the Michael Nielsens and Nadia Asparouhovas of the world), but I wonder who stands out for you as having done this well, and more on the lab/applied side.
Bret Victor comes to mind, though less of a researcher and more of a designer. I see murmurings here and there, but it needs explicit coordination to become a thing IMHO, and deliberate cultivation of these skill sets by a community of researchers. Might be worth putting together a workshop!
This is great. With regard to the tooling aspect, I think you should really look into existing approaches and where exactly they currently fall short and break. I'm thinking of efforts like PubPub (https://www.knowledgefutures.org/pubpub/) and OurResearch(https://ourresearch.org). Travis Rich's thesis (creator of PubPub) overlaps with a lot of what you're talking about so also worth checking out (https://www.media.mit.edu/publications/collaborative-scientific-publishing/).
This is a great tip — thank you.
My hunch (and hope) is that a lot of the most promising ideas here will start as integrations inside existing products/tools — extensions to arXiv, overlays on PDFs, plugins for Benchling/Notion/Slack — so that researchers can try them in situ, give early feedback, and we can see what actually sticks. Then, if something clearly works, it can evolve into a standalone layer.
I feel like we need a new job role in academia to just do all of this and manage it at department/school/faculty etc level. So it becomes as easy as updating a spreadsheet for the researcher/PI. Build a KPI around it etc. Then it is covered minimum level for all and those that want to excel and lean into this area can. Not all scientists will be comms stars and that's OK imo.
Some labs I’ve visited have a designated graphic designer that goes into this, which is fine when the teams are small, but then leads to backlogs and makes the smallest edits/changes a huge back and forth. Ultimately this becomes unsustainable IMO - increases friction, than if we just had simple plug and play tools that let everyone (researcher) control the narrative, look, tone, feel the way they like it.
It’s a bit like ghostwriting for tweets and using AI and stuff. Generally with writing, each person is uniquely different and it’s the little nuances that would benefit from having individual control over.
Great ideas!
This all has to start with a shift in the scientific culture. Right now, there are only two recognized currencies for academic scientists:
1. Grants funded
2. Peer-reviewed papers
So, there seems to be no good incentive to communicate with the public. Time spent in public engagement is time not spent writing grant applications, doing experiments, or writing papers — the very products that one currently needs for career advancement.
We need to start training grad students skills in public engagement, and ensure that universities also value it (i.e., in tenure and promotion decisions). The scientific ecosystem needs PIs to produce original research, but it also needs people who can effectively communicate it.
I recently wrote about this in Nature, for anyone interested: https://www.nature.com/articles/d41586-025-02258-7
What are some institutions that currently train their grad students well? Have you seen this done correctly?
I don't know the answer to this. I mostly work in the world of teaching clinical students (doctor of physical therapy), and have not been all that connected to typical PhD / post-doc training environments in recent years.
So, I think a typical grad program has room for some mentorship with communication. For example, the process is not done at the completion of a peer-review paper, but also extends to the next step of public engagment.
Conversely, I think most clinical programs (like DPT, MD, DVM, etc.) don't have as much flexibility to do this, even though it would be valuable. They are geared at creating entry-level clinicians, which involves preparing for a national-level licensing exam. So, the curriculum may be a bit more "rigid." When communication is emphasized, it's often based on communicating with individual-level patients, rather than broad public engagement.
One of the courses I teach touches on this, and includes an assignment in which they need to create a 3-5 minute answering a clinical question, aimed at a general audience. So we focus on select aspects, like ensuring the language isn't too technical (unnnecessary medical terminology, p-values, etc.), and trying to remain concise, yet accurate in their summary. It's not a comprehensive course on public engagment, but it's a start.
So, long answer, I don't know what grad programs are doing this well. Interested to hear others' responses.
Appreciate the thoughtfulness and candor here, thank you!
Was so excited for this after Part 1, and can’t wait to read this tonight!
Excited to hear your thoughts!
Thought provoking and engaging read.
I am not sure if I am on track where you mentioned of who is "experimenting not just with hypotheses, but also with documentation styles and incentive systems in public?" but Arcadia Science https://www.arcadiascience.com/ comes to mind in a not for profit but also non-academic mix pioneering new tooling, research and publication styles. I can't say there is a clear success or not, but an interesting experiment of how to create, perform, and communicate research.
> That convergence of tech (a vertical) and mainstream culture (horizontal) didn’t just happen because the code got better. It happened because the tooling to make it legible got better. Mac OS, GitHub, Figma, Twitter — each of them lowered the lift of distribution, making technical work culturally relevant — enough to catch the eye of even Anna Wintour
This is bullshit word salad. WTF does it even mean? macOS has been around for 40 years and I doubt the average Vogue employee or reader has any idea what GitHub or Figma is. Bezos caught Anna Wintour’s eye for the same reason Elon did: they’re the two richest men on the planet, not because she took a fancy for AWS.
Think you're missing the point here. The point I'm making is that there are 3 waves of tooling and distribution. Bezos and Elon caught attention because they're rich, but how did they get rich? 1) building a useful tool (the code, which I address), but also 2) knowing how to sell. And selling means knowing your user's pain points, marketing it to them well (or hiring a team that can).
For this to even have gotten to the attention of a Vogue employee means that tech did something right. For it to have crossed domains like this - outside of the world of code and software - was about legibility.
I don't think so. Amazon Studios is a major player in the Hollywood ecosystem so of course Vogue is going to suck up to him; plus Bezos loves hanging out with celebrities (see: launching Katy Perry into space).
Bill Gates, Larry Page, Sergey Brin, and Steve Jobs were never invited to the Met Gala, even though their products are more ubiquitous than Amazon and are likely used by every single Vogue employee every single day. And Jobs was 100x the salesman Bezos ever will be. Amazon minus AWS is not a tech company but Walmart on the web.
The Met Gala is only one example that emphasizes that tech has been so deeply embedded into culture/pop culture. I could reference Zuckerberg and having a film produced about him (in a somewhat negative light, so doubt he had a studio or financial influence over this), Apple having advertisements and endorsements with A-list celebrities, however medium you want to argue, the impressive thing is they’ve broken through in a way research hasn’t.
My point remains: the excerpt I quoted still makes no sense.