But is it really a strategy? Or is the entire foreign policy just based on ‘vibes’ from a person cult around delusional narcissistic guy who just incoherently rambles about whatever comes to his mind?

Calling hidden prompts for LLMs ‘bribery’ is overshoots the goal. If we want every paper to be “digestible” by an LLM, we “enable” LLM reviews. Then, there is an incentive to write (with meaningless changes or not) in a way that LLMs give favourable reviews to, instead of writing for humans.

Not sure if “bribing a reviewer” is the right comparison. Is it not literally more like asking for a good review?

But isn’t the real scientific misconduct reviewers who use LLMs for review? Of course hidden prompts in papers are questionable, but they don’t detract from the scientific results/progress like the opposing player (LLM reviews). Under normal circumstances, these prompts have no effect whatsoever

Here's the link to the quick and dirty script to do this: github.com/josauder/cor...

This has been a colossal effort from the team: data acquisition, 1500+ hours of expert annotation, and thorough benchmarking! 🤓

A big thank you to my co-authors @viktordo.bsky.social, Guilhem Banc-Prandi, Gabriela Perna, Anders Meibom, Devis Tuia, and all members and partners of the @trsc_org!

More details in the pre-print: arxiv.org/pdf/2503.20000

Coral reefs are under unprecedented threat, and monitoring is crucial to devising and assessing effective conservation strategies.

Automated coral surveys based on computer vision have potential to scale monitoring up!

Project website: josauder.github.io/coralscapes/, with code, pre-trained models, and the download.

Coralscapes is deliberately designed to follow the structure of Cityscapes, facilitating the use of state-of-the-art semantic segmentation neural networks in a challenging ecological domain!