Leave aside the sensationalized claim (the researchers told the llm to replicate and made sure it had all the tools to do so), and this is still quite remarkable. The LLM was able to plan three steps ahead to accomplish a goal using the agent framework it was given.
Unfortunately the researchers don’t publish the setup or even say what agent framework they used. The corresponding author’s page also just links to his google scholar profile. https://secsys.fudan.edu.cn/ym/listm.htm
I really wish conferences and journals would stop accepting papers that don’t have enough details included or linked to replicate them. It was the same when I briefly worked in academia around 2005, and I see that there’s a long way to go with the papers with code movement.
Yes, but most of the preprints are uploaded before being submitted for «print» (conference proceedings included in the notion of print even if electronic-only), so the influence of the review norms is non-zero on arXiv, too.
Leave aside the sensationalized claim (the researchers told the llm to replicate and made sure it had all the tools to do so), and this is still quite remarkable. The LLM was able to plan three steps ahead to accomplish a goal using the agent framework it was given.
Unfortunately the researchers don’t publish the setup or even say what agent framework they used. The corresponding author’s page also just links to his google scholar profile. https://secsys.fudan.edu.cn/ym/listm.htm
I really wish conferences and journals would stop accepting papers that don’t have enough details included or linked to replicate them. It was the same when I briefly worked in academia around 2005, and I see that there’s a long way to go with the papers with code movement.Isn’t arxiv.org a site where basically anyone can upload pre-prints?
Yep, in scanning the page I thought I saw it had been accepted somewhere. My bad.
Yes, but most of the preprints are uploaded before being submitted for «print» (conference proceedings included in the notion of print even if electronic-only), so the influence of the review norms is non-zero on arXiv, too.
Thanks for expanding! I see a lot of links to Arxiv but don’t have a handle really on how “reliable” the submissions are.
I’m not saying that peer review is necessarily a mark of quality.
The influence of review norms on Google is also non-zero, because some of the results are peer reviewed papers.