The attack on Ledger’s supply chain would not be limited to Asia. Reports are circulating about a Ledger device purchased a few weeks ago from an official European retailer that contains unauthorized hardware. With suspected cases now reported in Asia, South America, and Europe, concerns are growing regarding the geographic scope of the attack on Ledger’s supply chain.

ANOTHER ATTACK VIA HARDWARE WALLET Someone has acquired a company within Ledger’s official distribution network in Southeast Asia and, according to allegations, used the company to sell tampered devices. CryptoBilis operated in Malaysia, Indonesia, and the Philippines. Since yesterday, wallet drains linked to customers who purchased from them have emerged; on-chain estimates suggest over $80 million (Bitcoin, Ethereum, Tron), a figure not yet confirmed by Ledger. The most widely cited hypothesis is a supply chain attack: devices equipped with a hidden surveillance module (an LTE chip installed where the screen pad should be, connected to the device’s bus) that intercepts the seed phrase during setup. In this regard, former Mt. Gox CEO Mark Karpelès described receiving a device from Malaysia with precisely such a modification. In short, CryptoBilis was sold in March, and according to corporate records, since August 3, 100% of shares have been held by an individual named Jiaming, residing in Heilongjiang Province, China. The new owner, after acquiring CryptoBilis from the previous Malaysian owner, allegedly imposed a non-disclosure agreement preventing the former owner from announcing the sale until October 19. The attached image shows the former owner’s response to a post on CryptoBilis’s X profile, effectively acknowledging that the new management now controls the account and operations—and that he no longer has access. Ledger has requested CryptoBilis to halt all shipments and has warned customers who purchased devices within the last 90 days not to set up their devices—or to transfer funds to a new Ledger device with a new seed phrase. In summary: rather than hacking Ledger directly, this criminal group targeted the weakest link in the official supply chain and planted compromised devices. The investigation is ongoing. Stay vigilant. #crypto #Ledger

You can laugh and joke, but good old Jim gave us a massive buy signal back in early August, when BTC was around $63,000. Unique and unmatched

In recent days, the debate on artificial intelligence has taken an unexpected turn. Jacob Coxon, a researcher formerly at OpenAI and Anthropic, left Anthropic, laying out serious concerns about the future of AI in writing. Just days later, Dario Amodei, co-founder and CEO of Anthropic, publicly called for slowing down the race toward more advanced models. Sam Altman, CEO of OpenAI, moved in largely the same direction. Individually, these events might seem like just another debate on AI safety. But when considered alongside DeepSeek and Kimi—two of China’s most advanced models—the growing difficulty in justifying hundreds of billions in investments, and the increasing involvement of governments, they point to something much larger. I’ve attempted to synthesize all of this into this essay: technical risks, economic incentives, geopolitics, regulation, and potential market consequences—without catastrophism or adherence to any preconceived narrative. The fundamental question is simple: What happens if those developing AI begin to doubt that accelerating is still the right path—just as the entire industry has been built on the assumption that this race must never slow down? Link to the essay in the first comment 👇

Milei unilaterally decreed that Argentina’s minimum wage will be 219 euros per month—a level even below what business groups (“las cámaras empresarias”) proposed (around 270 euros per month). The minimum wage is far below the poverty line, which in the country stands at over 288 euros for a single adult. Long live freedom, bastards.

According to Moody’s Analytics, the top 10% of American earners now generate nearly half of all consumer spending in the United States. This extreme concentration helps explain why aggregate data continue to show surprisingly resilient demand, even as a much larger portion of the population faces mounting pressure. In my view, the key point is that this top decile is also the segment most exposed to fluctuations in stocks, funds, real estate, and other assets. As long as these assets continue to appreciate, perceived wealth rises, along with the willingness to spend. Seeing one’s wealth grow over time instills greater confidence to buy a home, travel, upgrade a car, or simply spend more. And this is where the problem arises. A sharp market correction would drastically reduce the “wealth effect.” Those who can still afford such expenditures might simply choose to delay them, becoming more cautious as they suddenly feel “less wealthy.” If nearly half of consumer spending depends on this group, a sudden drop in asset values risks rapidly transmitting to the real economy. The rest of the population is unlikely to compensate, as they are already heavily burdened by the cost of living, increasingly restricted access to credit, and a slowing labor market. This is also why a brutal correction on Wall Street today would not merely be a problem for investors—it could quickly become a problem for the entire U.S. economy. An American economy increasingly dependent on the wealthy—and, above all, on their continued perception of wealth.

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.

Bitcoin has just staged one of the most compelling displays of strength throughout the entire bear market. In my new article, I explain why, after months, the burden of proof has now shifted to the bears... https://t.co/bFyC5fsXVC

It’s thanks to thousands of traders like this girl that we had one of Bitcoin’s most memorable short squeezes last week. The irony is that she might even turn out to be right about the lower low, but after such a price action, it would be completely irrelevant—since she was almost certainly liquidated. And we must also acknowledge the objective responsibility of leading thousands of her followers to ruin (a habit these shitcoiners casually repeat). There’s nothing wrong with being bearish on Bitcoin, but suggesting to short sell at the lows of such a volatile asset as BTC borders on the grotesque. Regardless, here’s to ten more, a hundred more, a thousand more Layah Heilperns.

The U.S. Strategic Petroleum Reserve has fallen to approximately 293 million barrels, the lowest level since 1982. Trump is releasing 172 million barrels—nearly identical to Biden’s 180 million in 2022—but starting from a lower base due to incomplete replenishment of reserves. Meanwhile, the diesel crack spread (the margin a refiner earns by turning one barrel of crude into diesel) has surpassed $100, a historical high. In this context, crude withdrawals from the SPR have limited impact: the real bottleneck is refining capacity. How much longer can this imbalance last?