2) the brand value of high JIF journals critically relies on the perceived rigour and competitiveness of the selection process. If any random slop can dish out a SNELL paper using AI, then their brand value wouldn’t be compatible with their current business model.

1) I am questioning whether there is actually going to be more noise. Was the process of producing text/figures really rate limiting for fraudulent actors or will AI incentivize more bad actors? I have yet to see any evidence from computational/numerical fields where this could have happened earlier

I was thinking of recruitment rather than publishing. I’m not sure what the consequences will be for publishing. Simpler fraud doesn’t necessarily mean more fraudulent actors. AI-generated work in glam journals will kill their brand value. And we could just continue to focus on quality over quantity

Right, but if openness to technology could enable us to reap the benefits you described, shouldn’t we blame the people rather than the tools for not following through? And can’t we expect some self-correction in the coming years, as resistance to change becomes a competitive disadvantage?

Is it really the AI models that are creating the environment of mistrust or is it the reaction of the established system? Wouldn’t any large change to the interpretation of evaluation proxies/metrics have a similar effect?

after becoming aware of just how wrong they can be, generative AI have begun urging users not to talk to their friends at all and encourage them to ask their friends about something they both are very knowledgable in so they can understand how fallible the humans are.

[so AI may be bad but probably no worse than your average pundit and definitely not as bad as WSJ 😆] 3/n

But you could already do all of this years ago with more numerical data types, such as flow, proteomics, or metabolomics, where visual sanity checks aren’t available.

how do you stop image generating AI from being injected into some step of the audit trail? licensing/standardisation will kill innovation in instruments and post-processing.

Vendors would probably like everyone to replace everything with blockchain-enabled machines, but that would still means no postprocessing, no lab-built equipment and expensive certificstion of blockchains. My guess is that we will just see a deepening of the reproducibility crisis.

Max, searching with keywords is so 2021, these days you have chatGPT find it for you! (I’m kidding, this seems like a bad idea) help.openai.com/en/articles/...

My impression this year is that we still receive many genuinely written texts, both strong and weak. As for the signal, we can’t know how those not invited would have performed. We see similar correlations in pre/post interview ranking, but I do wonder how much of it is just self-fulfilling prophecy

How do you know whether it was a meaningful signal that you were picking up before? Are all letters crap or good now?