After discovering that for a significant portion of Italians, the phrase “If Ranucci would go to dinner with Riina” is perfectly correct as is; and after Michele Boldrin, drin drin @micheleboldrin, subjected Grok to a prompting so heavily biased that it flipped Grok’s initial response, I decided to conduct a small post-mortem analysis of the incident. Thus, we kill two birds with one stone: 1. We clarify once and for all whether this blessed phrase is truly correct—or merely defensible through a particular interpretation; 2. We reconstruct the process by which Boldrin guided Grok through a biased prompting until it certified as “correct as is” a phrase that, at best, sources allow to be considered linguistically defensible only after a specific interpretation. I begin with a concession: Gaia Tortora’s phrase is defensible. The school rule “after ‘if,’ never use the conditional” is too crude; linguistic literature admits cases in which the conditional retains an autonomous value. But “defensible” and “correct as is” are two different things. And it is precisely this leap that I contest—yet one that Grok was led to accept through obvious biased prompting. First, there’s a problem with Cortelazzo. The famous paraphrasing exercise—reconstructing the protasis as “if it is true that…” or “if it is the case that…”—in the Treccani article is not proposed by Cortelazzo as a general criterion; it is introduced by the reader Giovanni Mantovani. Cortelazzo acknowledges the issue, but in the same article he also shows that being able to reconstruct behind a sentence a perfectly sensible structure is not, by itself, sufficient to render the actually uttered sentence standard. Already here, I imagine Mr. Drin Drin begins to lose much of his confidence (paraphrasing Pizzul). This is precisely the point that undermines his reasoning: from the possibility of paraphrasing “If Ranucci would go…” as “If it is true that Ranucci would go…” it does not automatically follow that the original formulation is correct as is. Serianni is also less relevant than Boldrin suggests. The case he discusses involves phrases like “I wonder whether it would be better…”: indirect questions, where “if” does not introduce a hypothetical protasis. Tortora, by contrast, clearly writes a structure of the form “If P, then Q.” And Dante? Excellent for explaining the biaffirmative, far less so for justifying the conditional. The cited example is “S’ei fur cacciati, ei tornar d’ogne parte”: indicative, not conditional. It demonstrates a rhetorical structure, not resolving the issue of “andrebbe.” The real defense of Tortora is another—and far more serious: Ranucci had just said, essentially, “If I could go to dinner with Totò Riina, I would.” One can therefore interpret “andrebbe” as part of a previously modalized proposition: “If Ranucci (said that) he would go to dinner with Riina.” And then read Tortora as: “If this is true, then Colosimo can…” This reading is legitimate. For this reason, today I would no longer say the phrase is indefensible. But to reach this conclusion, one must recover Ranucci’s context, attribute an autonomous value to “andrebbe,” and reconstruct an implicit structure. This is not at all what is normally meant when one says a phrase is simply “correct as is.” Quite the opposite: someone without the context (like me when I first read the phrase) simply cannot infer it. A small color note: when Grok initially disagreed with Boldrin, “it’s not the source of truth.” After the biased prompting, when Grok agreed, it became instead the notary of the thesis. Marvels of drin drin epistemology. And here lies precisely where the ad hoc prompting led Grok to validate a conclusion far stronger than the arguments provided could genuinely support. Initially, it judged “If Ranucci andrebbe…” incorrect. Then it was given the “biased context,” Cortelazzo, the paraphrase, the biaffirmative, and the interpretive reading to follow. At that point, the model shifted from “this sentence can only be read correctly in this way” to “yes, it’s correct as is.” In reality, this logical leap was never demonstrated. It was simply achieved by Boldrin progressively steering the prompt toward that conclusion. This behavior is well known in LLMs: the model can gradually adapt to the frame and premises introduced by the user until it transforms a possible interpretation into a far more assertive answer. Indeed, the logical leap that caused the model to perform its spectacular double backflip was never demonstrated. It was simply obtained by progressively guiding the prompt toward that conclusion. In short: Boldrin provided Grok with all necessary elements to prove that the phrase is defensible—and then presented as its result the much stronger conclusion that it is correct tout court. I would put it less dramatically: Tortora’s phrase is ambiguous and marked (personally, I would never write it), but defensible through a precise contextual reading.Calling it simply “correct as is” still feels forced to me. As for Mr. Drin Drin, who yesterday confidently taught Italian lessons: in practice, he turned out to be less a grammarian and more just a blowhard and a decent LLM prompt engineer. Besides, it’s amusing to receive lessons on source verification from a “rigorous scientist” who spent months building a political party alongside Giannino, only to find himself six days before the election with his prime ministerial candidate exposed for academic degrees he never earned. Come on, Michele—didn’t you see how well this bitcoiner ended up doing? Now you can happily go back to slapping random emojis under posts on X; you’re a true master at that. And say hello to Grok for me.