I am thinking of not replacing but assisting human review. Say, a human reviewer is assigned 10 preprints to screen. 9 of them are greenlighted by AI, so the human will only briefly glance at them. Only one preprint is flagged by AI, so the human will spend more time on this one preprint. Saves time

Basically, if we are to trust AI with decisions such as helping to make a medical diagnosis based on image recognition in histology analysis, where life-or-death of a patient is at stake (and the field is moving towards this), then trusting AI with screening of preprints poses much smaller risk

it'd be interesting to see the current percentage of rejected bioRxiv submissions - probably not so large. As with any ML classification model, it is possible to adjust the threshold and send for human checks e.g. 10% or 5% of submissions, etc - depending on the required sensitivity/specificity

e.g. AI could flag questionable submissions for human checks, and if not flagged preprints can go straight online

There are cases like *screening* of preprints to establish that they are in general suitable for bioRxiv, which can be outsourced to responsible AI without too much damage

Finally, we propose a model in which chromatin is self-organising based on histone acetylation and nucleosome depletion. We hypothesise that active cis-regulatory elements may contact one another at or above a layer of acetylated nucleosomes.

Great development! I literally thought few days ago that this combination would make perfect sense

In a messaging mode, probably few minutes per day, but multiple times

Talk - you mean by voice? I noticed that in the voice mode it switches from version 5 to 4, and the difference is remarkable, so i stopped using it in voice mode

Looking good! From our lab, hello to your lab! :)

enabled now

Here is what ChatGPT replies to this :)
- That’s like using a calculator before learning arithmetic
- Teachers cannot assume students “inspect” the changes
- Overreliance creates long-term dependency
- It can quietly shift student voice
- Language learning can be harmed (removes productive struggle)

I guess the main question is, where is the boundary of helpful language polishing and unhelpful use of AI for academic writing. For me, this boundary is somewhere at the limit of capabilities of standard Word's grammar check or Grammarly's style check (or equivalent).