Thomas Wolgast

Building software to fix academic publishing. Visit for my concept how to encourage reviewing and reduce low-quality submissions.

[Thomas Wolgast @ThomasWolgast]

@_TarunKathuria I am deeply convinced that, within a year or so, quantity of publications won't mean anything anymore, and quality will be everything. So, good call IMO!

[Thomas Wolgast @ThomasWolgast]

@juanof9 There is so much potential when we move away from outdated standard PDF papers.

[Thomas Wolgast @ThomasWolgast]

@AlhamFikri I feel we have to start rejecting the bad notion of "already done". Very often, research is very valuable without being novel (especially in engineering). And often it is not valuable although it is novel.

[Thomas Wolgast @ThomasWolgast]

@Aaroth The review system needs an update. I talked to 10+ journal editors last year and they all said that finding reviewers is a growing problem and they don't have a solution...

[Thomas Wolgast @ThomasWolgast]

@mariusmosbach I did not experiment with AI reviews a lot, but doesn't it depend on the prompt? For example, if you tell the AI to explicitly focus on the bigger picture, would that fail?

[Thomas Wolgast @ThomasWolgast]

@mattzeunert Our publishing system would work drastically better just by using better software that was not created in 2001...

[Thomas Wolgast @ThomasWolgast]

@surendra_pandar Hot take: intelligence is not unidirectional. There is no "as smart as".

[Thomas Wolgast @ThomasWolgast]

@0x_joshio Did you find good ressources to learn Architecture? There are so many ways to learn about syntax or tools, but I never found something good for learning Architecture.

[Thomas Wolgast @ThomasWolgast]

@ross_prager I'm optimistic that it will happen automatically. 100+ publications/year will automatically raise slop suspicion. Either way, it probably makes sense to actively design the system in that regards.

[Thomas Wolgast @ThomasWolgast]

@mguindani At the same time, we will probably have less preliminary work/results in general, because AI will make it easier to make the final jump.

[Thomas Wolgast @ThomasWolgast]

@fmy2kec It depends. For a quick prototype or standard CRUD, no need to review extensively. For a serious long-term project with real users, absolutely make sure you understand the code base. In the end, it boils down to developers understanding when to dive deep and when not.

[Thomas Wolgast @ThomasWolgast]

@oliviawpy2023 Most important reviewing advice I read: review the paper the authors wrote, not the one you would have written. Don't remember where I found that though

[Thomas Wolgast @ThomasWolgast]

@Anonymanasensei @ReviewerBase Thanks! APCs are only paid for publications, not for submissions. And I'm afraid the reject and desk reject rates will rise in the next years...

[Thomas Wolgast @ThomasWolgast]

@Andrew_Akbashev @ReviewerBase The big risk that I see with payments only is that it will be a strong incentive to game the system with massive AI usage to create reviews and make money.

[Thomas Wolgast @ThomasWolgast]

@Andrew_Akbashev Review incentives are one problem, but not the only one. @ReviewerBase is a general proposal for a non-profit reviewer database across the publishing system. It will contain reviewer expertise, review quality, etc, and a credit system (earn with reviews, pay for submissions).

[Thomas Wolgast @ThomasWolgast]

@IvanZupic Agree! AI mostly exposed existing flaws. Doesn't mean we can ignore the issues though. We have to redesign the system continuously to maximize benefit for humanity.

[Thomas Wolgast @ThomasWolgast]

@nielstron When I don't understand the paper, I usually stop the review and recommend reject or major revision. I also clearly communicate exactly that.

[Thomas Wolgast @ThomasWolgast]

@Michael_D_Moor I truly believe that high quantity of publications won't look impressive anymore. The opposite will be true.

[Thomas Wolgast @ThomasWolgast]

@Mehradity @GurushaJuneja But I also believe these are not unsolvable problems. For example, dividing paper citations by the number of authors would easily solve the second issue. It would just be a new metric.