Extra stuff as I get dm’s; will go into book in 3-12 months; expect these to be more messy than the post above
1: Masters,PHD’s etc isn’t a very strong indicator of skill level for AI ; like- they probably aren’t able to implement a cutting edge paper with no code (even in the same subfield); they might struggle to implement a paper with code,too.Can’t say much about other fields ; this was my bad I knew the skill levels were vague but didn’t realize how mixed of an indicator it has become
2:humanities is in such a “strange” spot
jreg has more impact than most phd’s—not just in the sense he’s a political pundit for zoomers, but he has created new research ideas/topics
but besides that- there’s a weird vibe in social science of using circular logic to prove things (not all the time obviously)
I think it’s also just that polsci especially is very much “believe whatever you want”, arguments is a glorified game of chess- the more you know the book moves, the more you can counter
as for like- psych? there is def some groups doing online stuff here, especially anyone looking into children of the internet stuff
more rare; it’s both walled off by academia, but also sometimes misrepresented via video essayists
(i.e- picking papers to fit an agenda; rather than using papers as a datapoint- which goes back to arguments and counters being a glorified game of chess)
as for philosophy
I’m going to be honest,I cannot,for the life of me,figure out what philosophy is
other than it being values and views on things from a bunch of old dead people
it does feel like a lot of kids and teens get into philosophy cause its some kind of mythical subject
at least, like 10 years ago, but i think it’s also because a lot of stuff that im assuming fall under philosophy i’d place into other fields eg:
rationalism and accelerationism etc is cultural (social?) theory
bias spotting? metascience and psychology memetics? - it’s its own field now,(something relating to psyops or something)
by this definition of philosophy- it does feel like if theres more non academics producing useful works than academics
I think this is also partially cause of the replication crisis
impact factors are lower than described, and sometimes inversed or no impact at all especially because these things are non linear something important to ask yourself, what shapes what people research?
it feels like more popular ideas tend to get researched more; even if the ideas aren’t from academia; though this might be bias as i haven’t fully looked into it
3: having juniors do good first issues for a github project of a research tool is nice, but I think with the rise of claude etc; the skill floor has increased so a lot of good first issues probably require you to spend like a week or two looking at the codebase,which might scare people off because of reasons i mentioned somewhere in the post.Google summer of code fixed this issue by having a few weeks to talk to the mentor,look at the codebase,
4:DAO’s can be swapped with the word desci and not too much changes but Desci is to opensci as the labs are to DAO’S
desci is the larger landscape for the crypto bros, opensci labs is the larger landscape for the independent research labs
5: yes,publishing negative results somewhere is good.
6:scicomm issues run a lot deeper with scientists just being busy etc
7:paper lengths could be both longer and shorter; maybe more meta analysis?
8: Academic life in undergrad vs grad school is very different; besides that I do think theres a lot of important info that just isn’t written down anywhere which people might think is a breakthrough (it relates to novelty of a paper)
9: Who an expert is, is definetly subfield dependent and how niche you want to go as well as your applications- general math? a professor, a specific subfield? someone who studied that subfield. A specific technique/paper in that subfield? see who did meta-analysis of the subfield/who wrote the paper, as well as related stuff.
10: community awards and interviews are nice too.
11: hard science (and engi) costs can go much lower with low cost science tools, or even just having a local fab shop thingy that does it in like 30 mins; supposedly china is setup like that
12: yes, you’re going to be stuck in internship hell. Or even volunteer hell. I don’t know where the money has gone.
13:as much as desci and opensci can be seen as two factions its a lot more complicated than that
14: outputs matter a lot in open sci labs—im not going to rank a group that puts out courses (or does like resume reviews etc) highly for example (because thats not really an opensci lab- closer to a study hub/education server) - if they make arxiv papers they’ll be much higher. Projects dying is the norm. I’ve listed some to stop this but I think theres a lot of issues at play just cause of the fact people join a server,look at it,then never look at it again.
15:the profitability of a research paper is anywhere between $300 to $30 000 . the median (all the way up to 80%?) paper brings in $0 to negative loss if you account for opportunity cost/cost of living etc I got the numbers based on openai profitability vs all arxiv papers as well as drug development costs + all drug development failures
16: soulbound tokens are very useful and can help with funding, same with smart contracts
17: to go into a bit more detail about teaching issues ; mix-maxing stuff that are good on a resume/CV isn’t really smart for research ; its not so much a critical thinking problem but like- people don’t know how to explore—especially because of strong foundations required to explore (otherwise you have ivy league highschool prospects doing ISEF etc which is good, but doesn’t create a culture of science) (sometimes it does ; but idt they really have good research taste or really know what they want to do yet- so they usually pick something from others which isn’t bad, but they’d probably need to read papers etc?) maybe rewrite this point...
18: Everyone doing their postdoc/phd research on their subfield is kind of an expert on their whatever their researching and usually nothing more; but people doing similar things can also be experts in that in their own subfields if they’re doing similar stuff eg: testing why an AI can turn toxic when fed certain images- each person working on this has their own theories and are essentially an expert on their own theory; but that doesn’t mean their work is fool-proof; other experts etc can still find flaws in it , and even the public depending on what was done (notable in softsci);I should rewrite this as well
19: Theres a lot of low hanging research fruits in AI- except most of them are like dataset and easier stuff to do; most of the advanced stuff are also higher impact. thats not to say theres high impact low hanging stuff to do; just most of the simpler low hanging stuff is probably not going to be too impactful compared to lets say a breakthrough in efficiency or something.
20: I was planning to make another sheet of skill levels but its really hard to gauge skill levels- everyone can call themselves an expert mostly cause they think they’re an expert; after all, skill improvement always goes up; you’re always at the peak.Good research taste is probably a better indicator of expertness- but not so much indicator how well you can implement things
21: as with most groups/companies—if they get popular enough (or impact to certain people) - you can start getting talks with ceos, researchers (researchers always like talking about their work) invited to conferences,etc.
22: research taste and tatic knowledge are probably the most useful skills for a scientist (next to like- being able to implement stuff; maybe its cause ai helps a lot with implementations now).
23: most current PI’s etc prob don’t see too much faith in this; give it like 5 more years or something; it’s already here for compsci, will be better in the informatic fields etc soon,
24: various fields cost differing amounts to be upskilled into, math is 0,compsci is 0 (excluding opportunity cost,wet lab sci is much higher) - ill go more into this in the book
25: skills can be probably be learnt much quicker than most would imagine- i’d say from finished high school to starting PhD could be done in like 3 years? especially with AI to help; i’d go more into this in the book, using math as an example for a syllabus to doing your own research.
26: pre-2022 or so,when AI was but a cool niche, there were both less servers, and most of them were really technical. (at least those that didn’t talk about AI as sci-fi) , I think sometime since 2025; maybe because of the bubble? a lot of the pre-2022 or so researchers left these spots to work in industry, run a startup,etc. The new researchers that filled in aren’t as good as the last group; this is continuous but I think because AI is mainstream a lot of even mid tier scientists, or really anyone okay with AI, now having a pretty good job,Ig it relates to “problems” with fields going mainstream. I’m sure if some subfield of math turned mainstream the same thing would happen. these groups were always meant to be a stepping stone
27: I talked about this a bit with the crackpot problem,but making a place more accessible brings more crackpots and newbies than it ever could bring experts, and if those newbies are very active quality drops massively and experts move somewhere else.
28: related to culture building but like- a lot of the more practical stuff like talking to professors to do research in undergrad/hs is both not possible in most third world (since no real research unis), as well as no metacognition to think through those thoughts
29: I’ll go into more detail in a bit but theres a lot of important stuff you should be able to think through eg: questions to consider to find hidden implications, conclusions that should work based on what is the theory, flaws in the study; but not just in study but in writing in general, akin to a deposition to poke holes at an argument?- i’ll make a checklist of questions soon.
30: but not just those: theres also another set of questions relating to like- how to relate this to other things?is there an easier way to do this?can you use this in a unique way? maybe inverse things and see what happens? it helps both with solidfying what you know , helps with searching for problems,Or just understand why things are done the way they were done (if there is any reasons;chestersons fence and all that- getting the reason can solidfy the theory or create new experiments if they just tried it that way since thats what they knew) i will also, make a checklist of questions for this too, soon
32: I think this also kinda explains why compsci is so popular- the high amount of reusing other code probably would be called plagarism depending on the field- besides the ability to reuse code like that brings you to about diploma or masters level?
33: I think theres also a shift from the user end 2013-2022 era FANG developers: certs,github,projects,go to uni 2023-now AI devs: Reading groups,publish your own papers,fellowships,don’t go to uni.
34: the reason why AI and compsci took off compared to other fields is also partially just that other fields don’t have an easy ramp or PhD equivalent eg: for humanities you can do activism,consultancy, policy making but none of which give you research skills; research skills and skill in AI are decently intertwined- especially if you’re just trying stuff or implementing cutting edge stuff rather than just doing practice/applied (think back to the example of the doctor vs pharmacologist)
[image showing someone gaining research skills doing compsci here while the help out others in other fields doing practice here,akin to an exp bar]
Extra stuff as I get dm’s; will go into book in 3-12 months; expect these to be more messy than the post above
1: Masters,PHD’s etc isn’t a very strong indicator of skill level for AI ; like- they probably aren’t able to implement a cutting edge paper with no code (even in the same subfield); they might struggle to implement a paper with code,too.Can’t say much about other fields ; this was my bad I knew the skill levels were vague but didn’t realize how mixed of an indicator it has become
2:humanities is in such a “strange” spot
jreg has more impact than most phd’s—not just in the sense he’s a political pundit for zoomers, but he has created new research ideas/topics
but besides that- there’s a weird vibe in social science of using circular logic to prove things (not all the time obviously)
sci-comm,like real research sciomm, helps https://x.com/joliegans/status/1964062256561619025
I think it’s also just that polsci especially is very much “believe whatever you want”, arguments is a glorified game of chess- the more you know the book moves, the more you can counter
as for like- psych? there is def some groups doing online stuff here, especially anyone looking into children of the internet stuff
more rare; it’s both walled off by academia, but also sometimes misrepresented via video essayists
(i.e- picking papers to fit an agenda; rather than using papers as a datapoint- which goes back to arguments and counters being a glorified game of chess)
as for philosophy
I’m going to be honest,I cannot,for the life of me,figure out what philosophy is
other than it being values and views on things from a bunch of old dead people
it does feel like a lot of kids and teens get into philosophy cause its some kind of mythical subject
at least, like 10 years ago, but i think it’s also because a lot of stuff that im assuming fall under philosophy i’d place into other fields eg:
rationalism and accelerationism etc is cultural (social?) theory
bias spotting? metascience and psychology memetics? - it’s its own field now,(something relating to psyops or something)
by this definition of philosophy- it does feel like if theres more non academics producing useful works than academics
I think this is also partially cause of the replication crisis
impact factors are lower than described, and sometimes inversed or no impact at all especially because these things are non linear
something important to ask yourself, what shapes what people research?
it feels like more popular ideas tend to get researched more; even if the ideas aren’t from academia; though this might be bias as i haven’t fully looked into it
3: having juniors do good first issues for a github project of a research tool is nice, but I think with the rise of claude etc; the skill floor has increased so a lot of good first issues probably require you to spend like a week or two looking at the codebase,which might scare people off because of reasons i mentioned somewhere in the post.Google summer of code fixed this issue by having a few weeks to talk to the mentor,look at the codebase,
4:DAO’s can be swapped with the word desci and not too much changes but Desci is to opensci as the labs are to DAO’S
desci is the larger landscape for the crypto bros, opensci labs is the larger landscape for the independent research labs
5: yes,publishing negative results somewhere is good.
6:scicomm issues run a lot deeper with scientists just being busy etc
7:paper lengths could be both longer and shorter; maybe more meta analysis?
8: Academic life in undergrad vs grad school is very different; besides that I do think theres a lot of important info that just isn’t written down anywhere which people might think is a breakthrough (it relates to novelty of a paper)
9: Who an expert is, is definetly subfield dependent and how niche you want to go as well as your applications- general math? a professor, a specific subfield? someone who studied that subfield. A specific technique/paper in that subfield? see who did meta-analysis of the subfield/who wrote the paper, as well as related stuff.
10: community awards and interviews are nice too.
11: hard science (and engi) costs can go much lower with low cost science tools, or even just having a local fab shop thingy that does it in like 30 mins; supposedly china is setup like that
12: yes, you’re going to be stuck in internship hell. Or even volunteer hell. I don’t know where the money has gone.
13:as much as desci and opensci can be seen as two factions its a lot more complicated than that
14: outputs matter a lot in open sci labs—im not going to rank a group that puts out courses (or does like resume reviews etc) highly for example (because thats not really an opensci lab- closer to a study hub/education server) - if they make arxiv papers they’ll be much higher. Projects dying is the norm. I’ve listed some to stop this but I think theres a lot of issues at play just cause of the fact people join a server,look at it,then never look at it again.
15:the profitability of a research paper is anywhere between $300 to $30 000 .
the median (all the way up to 80%?) paper brings in $0 to negative loss if you account for opportunity cost/cost of living etc
I got the numbers based on openai profitability vs all arxiv papers as well as drug development costs + all drug development failures
16: soulbound tokens are very useful and can help with funding, same with smart contracts
17: to go into a bit more detail about teaching issues ; mix-maxing stuff that are good on a resume/CV isn’t really smart for research ; its not so much a critical thinking problem but like- people don’t know how to explore—especially because of strong foundations required to explore (otherwise you have ivy league highschool prospects doing ISEF etc which is good, but doesn’t create a culture of science) (sometimes it does ; but idt they really have good research taste or really know what they want to do yet- so they usually pick something from others which isn’t bad, but they’d probably need to read papers etc?) maybe rewrite this point...
18: Everyone doing their postdoc/phd research on their subfield is kind of an expert on their whatever their researching and usually nothing more; but people doing similar things can also be experts in that in their own subfields if they’re doing similar stuff eg: testing why an AI can turn toxic when fed certain images- each person working on this has their own theories and are essentially an expert on their own theory; but that doesn’t mean their work is fool-proof; other experts etc can still find flaws in it , and even the public depending on what was done (notable in softsci);I should rewrite this as well
19: Theres a lot of low hanging research fruits in AI- except most of them are like dataset and easier stuff to do; most of the advanced stuff are also higher impact. thats not to say theres high impact low hanging stuff to do; just most of the simpler low hanging stuff is probably not going to be too impactful compared to lets say a breakthrough in efficiency or something.
20: I was planning to make another sheet of skill levels but its really hard to gauge skill levels- everyone can call themselves an expert mostly cause they think they’re an expert; after all, skill improvement always goes up; you’re always at the peak.Good research taste is probably a better indicator of expertness- but not so much indicator how well you can implement things
21: as with most groups/companies—if they get popular enough (or impact to certain people) - you can start getting talks with ceos, researchers (researchers always like talking about their work) invited to conferences,etc.
22: research taste and tatic knowledge are probably the most useful skills for a scientist (next to like- being able to implement stuff; maybe its cause ai helps a lot with implementations now).
23: most current PI’s etc prob don’t see too much faith in this; give it like 5 more years or something; it’s already here for compsci, will be better in the informatic fields etc soon,
24: various fields cost differing amounts to be upskilled into, math is 0,compsci is 0 (excluding opportunity cost,wet lab sci is much higher) - ill go more into this in the book
25: skills can be probably be learnt much quicker than most would imagine- i’d say from finished high school to starting PhD could be done in like 3 years? especially with AI to help; i’d go more into this in the book, using math as an example for a syllabus to doing your own research.
26: pre-2022 or so,when AI was but a cool niche, there were both less servers, and most of them were really technical. (at least those that didn’t talk about AI as sci-fi) , I think sometime since 2025; maybe because of the bubble? a lot of the pre-2022 or so researchers left these spots to work in industry, run a startup,etc. The new researchers that filled in aren’t as good as the last group; this is continuous but I think because AI is mainstream a lot of even mid tier scientists, or really anyone okay with AI, now having a pretty good job,Ig it relates to “problems” with fields going mainstream. I’m sure if some subfield of math turned mainstream the same thing would happen. these groups were always meant to be a stepping stone
27: I talked about this a bit with the crackpot problem,but making a place more accessible brings more crackpots and newbies than it ever could bring experts, and if those newbies are very active quality drops massively and experts move somewhere else.
28: related to culture building but like- a lot of the more practical stuff like talking to professors to do research in undergrad/hs is both not possible in most third world (since no real research unis), as well as no metacognition to think through those thoughts
29: I’ll go into more detail in a bit but theres a lot of important stuff you should be able to think through eg: questions to consider to find hidden implications, conclusions that should work based on what is the theory, flaws in the study; but not just in study but in writing in general, akin to a deposition to poke holes at an argument?- i’ll make a checklist of questions soon.
30: but not just those: theres also another set of questions relating to like- how to relate this to other things?is there an easier way to do this?can you use this in a unique way? maybe inverse things and see what happens? it helps both with solidfying what you know , helps with searching for problems,Or just understand why things are done the way they were done (if there is any reasons;chestersons fence and all that- getting the reason can solidfy the theory or create new experiments if they just tried it that way since thats what they knew) i will also, make a checklist of questions for this too, soon
31: https://www.tspi.at/2025/10/13/whatisscience.html#gsc.tab=0
32: I think this also kinda explains why compsci is so popular- the high amount of reusing other code probably would be called plagarism depending on the field- besides the ability to reuse code like that brings you to about diploma or masters level?
33: I think theres also a shift from the user end 2013-2022 era FANG developers: certs,github,projects,go to uni 2023-now AI devs: Reading groups,publish your own papers,fellowships,don’t go to uni.
34: the reason why AI and compsci took off compared to other fields is also partially just that other fields don’t have an easy ramp or PhD equivalent eg: for humanities you can do activism,consultancy, policy making but none of which give you research skills; research skills and skill in AI are decently intertwined- especially if you’re just trying stuff or implementing cutting edge stuff rather than just doing practice/applied (think back to the example of the doctor vs pharmacologist)
[image showing someone gaining research skills doing compsci here while the help out others in other fields doing practice here,akin to an exp bar]