Raphael Wimmer

Ja, wenn sich jedes Virus seine Saisonalität selbst aussuchen kann - gibt's denn überhaupt Evidenz , dass der Winter irgendwas treibt?

...

This occupied my mind more than your observations and predictions - which I find quite relatable and plausible.

...
4. oh, lots of text - I wonder how much of it was generated or expanded by an AI - let's ask Claude to summarize it for me
5. nah, of course not - it feels rude to AI-summarize blog posts written by people I know
6. if it's AI-expanded, why not use AI to reduce it to its essence again?

...

My reading experience:

1. oh, sound's interesting - let's check it out
2. oh, a picture of a whiteboard - let's check out the author's thought process
3. oh, AI-generated, it is impossible for me to discern whether a detail on the there is intentional or probabilistic - let's skip it :(
...

One more thing: We could get rid of many of the negative aspects of LLMs for academic publishing by:

a) not coupling careers to publication count
b) moving to a post-publication peer review model where quality assurance is provided by a web of trust instead of the prestige of a journal.

point of view than before. LLMs make it easier not to think long and hard. And that's dangerous for science and society.

In discussions about LLMs, People like to emphasize that "writing is thinking", and this is true in my case. I spent really too much time writing these few posts because I had to collect and filter my thoughts. I wrote many partial sentences that I then deleted. And I am less sure about my

Sorry for the long-winded reply.
tl;dr: LLMs are qualitatively different from other tools because they are able to take over agency over the research process. And I think that enough people in academia are vulnerable or short-sighted enough to abuse them.

Most research tools are inherently "lean-forward"¹ media - you need to actively engage with them.
LLMS are "lean-back" media - they actively encourage you to be passive while they do your bidding.

¹) www.nngroup.com/articles/wri...

LLMs are like a rock that you push down an incline in the jungle. Minimal effort, and it clears a perfect path. But steering the rock in the direction that you want is much harder than just following it in the direction that it rolls by itself.

Regarding "LLMs are just a tool like the internet": yes and no. A search engine is like a machete that allows you to explore the jungle of knowledge. It makes you 10x faster. But effort and result are still in a linear relationship - the more time you spend, the more you find.
...

... but you have learned that the LLM saves you from burning out. It takes away your sorrows.

Sure, there are lots of great research groups where this would never happen. But I am not sure anymore whether they are even in the majority.
Academia does not always select for empathy and diligence.

... skim your paper. It is a little bit boring, but there are no visible signs of your struggles in the draft because the LLM makes everything smooth. 'Looks okay', they say. You submit the paper. It gets rejected or accepted. You have not learned to write better. ...

Your PhD advisor might not have time this week - but the LLM is available whenever you need it. And what it tells you sounds reasonable. Your advisor wants to read the paper draft tomorrow, but it is still a mess. Let's ask the LLM for help. Your advisor is busy and only has half an hour to ...

In my field (human-computer interaction), PhD students are expected to publish multiple papers at conferences/journals with acceptance rates between 20% and 40%. It is a numbers game. In other fields, there is even more pressure to publish - and to publish faster than others.

Who writes the majority of papers? Typically young PhD students, I think. Senior scientists often only get involved in the writing process when major parts of a paper have already been written. (Correct me if I'm wrong - that's my observation across a few disciplines.)

Thanks for the thoughtful response!
These small character limits make it hard to write nuanced posts - and I posted my raw thoughts late in the evening after playing a bit with Prism. That said, I *do* think that easy access to LLMs in your writing tool makes for a slippery slope.

c) A reader who has seen my post or sees it later on might be negatively biased towards me by your framing.
d) I don't think posting rage-baity comments does contribute to a constructive debate on an important topic.

Have a nice day.

Hi. While I share your general sentiment:

a) You are using stuff I wrote - and build your argument on it - without attribution.
b) A reader who hasn't seen my post yet might believe that there are actually people who'd be proud to clog scientific publishing. I have seen no such evidence.

People with a pro-science bias should be banned from any and all advisory panels.

Thank you for the reply!
I kind of agree with you on the technical aspect.
My problem with the 'parrot' term is that it feels anthropomorphizing (for lack of a better term) to me.
Parrots are these cute birds who talk to humans in order to be not alone. That's not how I perceive LLMs :)

I'll let my AI assistant call the editor every day, asking them to fast-track my paper, please.
For every paper they reject, we'll send two more. If I can't get my paper published, nobody else should either.

Why won't ChatGPT tell me how to build a bomb but Prism will happily fabricate fake experimental results for me?

I am a rather slow writer who might profit from an AI writing aid.
A good tool would encourage me, help me while *I* am writing, and maybe keep me from taking shortcuts.

Prism does none of these things - instead it pushes me towards sloppy practices.

We’re getting so many journal submissions from people who think ‘it kinda works’ is the standard to aim for.

On the Dangers of Stochastic Parrots | Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency

The term has been coined in this paper: dl.acm.org/doi/10.1145/...

I actually don't like it that much because it minimizes how powerful current LLMs actually are. (Both the good and bad kind of power.) They are not simply word-completion machines but train-of-thought completion machines IMHO.

Bilbo looking at his phone top on bottom is ChatGPT After all, why not? Why shouldn't I keep it? You're absolutely right — you found it, it's been with you a long while, and it's only natural to feel fond of something that's served you so well, especially when someone like Gandalf suddenly seems to want it for himself.

Yes, we already have enough systemic problems in science. Unfortunately, AI does not solve any of them but creates new ones or amplifies existing ones.

Peer review science relies, more than we care to admit, on what @zey.bsky.social calls "load bearing frictions" that make it difficult to fake good science, and on trust that most ppl also value science itself & won't cheat just for personal gain. This creates the wrong environment for that to hold.

I guess that's my point, too.