Dan Goodman on Nostr: I think the two key words to notice in the title are "imagine" and "reliable". ...

I think the two key words to notice in the title are "imagine" and "reliable". Indeed, if you look in the paper the word reliable only appears 3 times in the text, and only once substantively. What they mean by it is that the agent regenerates the figures of the paper (after iterating with the goal of doing so). As far as I can tell they didn't check if it used the correct methods to generate those figures except for a handful of case studies. It managed to do this figure generation in 74% of their tests, and then answered a standard set of reference questions (so presumably quite generic) with 90% accuracy. I will leave you to decide if you think this meets your own personal criteria for reliability.