Veli‐Matti Karhulahti

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 1d

Hands down the most genuine, reflexive, & sensible take on AI as far as my memory goes-- this is canon

Julian Togelius @togelius.bsky.social · 1d

was at an event on AI for science yesterday, a panel discussion here at NeurIPS. The panelists discussed how they plan to replace humans at all levels in the scientific process. So I stood up and protested that what they are doing is evil.

Full post:
togelius.blogspot.com/2025/12/plea...

Please, don't automate science!

I was at an event on AI for science yesterday, a panel discussion here at NeurIPS. The panelists discussed how they plan to replace humans a...

togelius.blogspot.com

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 1d

I think they mentioned author email was traced back to the journal so sounds like the journal was trying to boost its IF by creating a highly cited meta

Julian Togelius @togelius.bsky.social · 1d

was at an event on AI for science yesterday, a panel discussion here at NeurIPS. The panelists discussed how they plan to replace humans at all levels in the scientific process. So I stood up and protested that what they are doing is evil.

Full post:
togelius.blogspot.com/2025/12/plea...

Please, don't automate science!

I was at an event on AI for science yesterday, a panel discussion here at NeurIPS. The panelists discussed how they plan to replace humans a...

togelius.blogspot.com

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 2d

asked the same question a year ago and ppl I trust recommended this, didn't read yet but next in line

mitpress.mit.edu/978026204880...

The Blind Spot

“This is by far the best book I've read this year.”—Michael Pollan, Professor of the Practice of Non-fiction, Harvard University; #1 New York Times bes...

mitpress.mit.edu

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 3d

2) evaluating in-depth qual/mixed requires expertise & knowledge and there's a natural evolution of OS folks being more quant oriented, many places simply don't have the competence to understand what such papers are doing so i'm also sympathetic to their allergies

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 3d

there's at least two major factors in it: 1) prestige publishing needs citable claims to maintain IF and public attention so not all epistemic efforts count

bsky.app/profile/mkar...

Veli-Matti Karhulahti @mkarhulahti.bsky.social · Oct 5

Major journals using statistics as a proxy for impact & quality these days nicely reflects academia's attention economy at the cost of being interested in explaining things

Joe Bak-Coleman @jbakcoleman.bsky.social · Oct 5

My hottest academic take is that we shouldn’t be using statistics in the vast majority of papers.

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 3d

It's a longer topic (& wrote them about it too) but can very much assure getting an in-depth qual/mixed paper published in a prestige journal almost never happens bc editors see them unable to make claims ("not enough N") and it's often framed as not meeting high OS

jyx.jyu.fi/jyx/Record/j...

Registered reports for qualitative research :: JYX

jyx.jyu.fi

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 3d

Yes but the point is that such rules are frequently weaponised against mixed/qual in particular as justification for their lack of credibility implicitly & explicitly--

remember when NHB initiated RRs? they made it clear from the start that mixed/qual wasn't welcome (and afaik still isn't) etc etc

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 3d

Here's example from a few years ago, rejection quote from major psych journal chief editor-- our qual paper wasn't able to meet mandatory raw data sharing (mental health interviews)

the paper fails to follow the submission guidelines which 2020, require a Data Sharing Statement as described. The article should have included the complete raw data-set collected in the study including the

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 4d

you can share pseudonymous data with collaborators (& reviewers) *as anonymous* when sharing is consented-- this can make things easier with some countries where transfer was previously a challenge, applies retroactively too if the term "anonymous" has been used

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 4d

in theory, seems you could now share pseudonymous data publicly as long as you keep the key and participants are informed of anonymous data sharing, the data isn't classified personal anymore, moreover--

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 4d

Been thinking about the practical implications of the recent EU court of justice ruling where data are now defined by the recipient: even if the data is pseudonymous for you, it can be anonymous for *others* if they don't hold decryption key

curia.europa.eu/jcms/upload/...

curia.europa.eu

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 5d

Confused about the moral of this story

Juno Rylee Schultz @junoryleejournalism.com · 5d

This bird’s greed delights me

Nick Huntington-Klein @nickchk.com · 5d

last night my daughter asked whether there were negative letters, like negative numbers. what a good fuckin question

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 5d

I see improving quality-control & thoughtfulness complementary and parallel processes-- both important but the latter more critical (science as a strong link problem) while also damn difficult and slow to deal with

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 5d

thoughtful work could be incentivised too but that'd require acknowledging prestige or sorts (not all studies are equal, some more interesting)-- this has been a red flag for a long time; lacking conceptual & theoretical attention, prestige remains unquantifiable while implicitly used all the time

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 5d

the reason why i've chosen to invest large parts of my volunteer labor on registered reports is exactly this: it's one of the few existing tools that can actually (sometimes, not always) fix broken designs-- reducing shit is even more difficult, as it's a systematic incentive problem

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 5d

This is the most worrying part the current system and the #1 issue to be tackled with our remaining brain cells, what can be done to a) reduce shit, b) transform shit into more thoughtful work

Darren Dahly @statsepi.bsky.social · 5d

A key theme from the talk:

Juno Rylee Schultz @junoryleejournalism.com · 5d

This bird’s greed delights me

Darren Dahly @statsepi.bsky.social · 5d

A key theme from the talk:

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 6d

high-level quality control doesn't produce prestige as utilised in the field & ongoing dominance of prestige for-profit publishers is partly due many (valuable as such) renovations focusing most efforts on quality control-- the prestige market remains essentially unchallenged

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 6d

Agree with many things in the talk & communicating this with detail is valuable work but picking a small thing that many repeatedly overlook imo: quality control does *not* equal prestige-- journal prestige is generated by a process of selection for impact

journal prestige can and should be earned

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 7d

My experience is similar whenever i've checked PLOS peer reviews, it rarely seems high-level (and let's face it none really is)-- yet again at least it's possible to check which is a feature all journals have (never published there, the APC is too high for my uni)

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 7d

Would agree that PLOS has a volume problem too but there's one major difference-- it runs open peer review so its editorial work is generally assessable

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 8d

apparently the JUFO index is used also in other countries like Slovakia, Denmark, Iceland etc so community action like this can snowball higher impact

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 8d

next year we're moving to a 2-level system (1= normal 2=prestige) & the call for community proposals is out-- if enough ppl submit justifications for diamonds they can take over some for-profits

julkaisufoorumi.fi/en/news/prop...

Propose changes to the level classifications of publication channels

julkaisufoorumi.fi

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 8d

Very curious to see the data by end of 2025 but after the first MDPI de-indexing cases in 2022 there was drop from 2800->2400, this year (and next) it should be much more

julkaisufoorumi.fi/en/news/publ...

Publication volume decreased in MDPI journals

julkaisufoorumi.fi

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 8d

problem 2: so far it hasn't stopped anything but MDPI/frontiers-- even tho 2026 comes with some diamond incentives, big nature/elsevier journals are still 'prestige' ie ppl prefer them over low-IF diamond --> IMO a missing but relatively easy solution is more non-profit prestige markers

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 8d

problem 1: it's "relatively stable" as it's very heavy to run an index with all journals, it's tons of manual work and definitely includes some errors-- self-correcting in practice but also vulnerable to fluctuating expertise

Veli-Matti Karhulahti @mkarhulahti.bsky.social · 8d

Re a few separate things in this thread: we're doing national indexing in FI and ended up very differently = almost all MDPI and frontiers are now banned, submission rates seem to approach zero (waiting for data end of this year) and the system is relatively stable

Add to Home Screen

Light up
your news

Add to Home Screen

Light upyour news

Sign in to Lightnews

Sign up to start reading

Connect Bluesky

Connect with Bluesky

Light up
your news