The failure analysis in First Proof’s Appendix A describes something qualitatively different from the hallucination patterns studied in factual QA: models producing proofs that are fluently wrong, where the wrongness is concentrated in a small number of unjustified load-bearing claims rather than spread across obviously false individual facts. I have tried in this paper to give that pattern a precise enough description to be studied systematically. The taxonomy has four modes (F1: citation fabrication, F2: premise smuggling, F3: silent reformulation, F4: local-to-global gap), and my empirical audit of eight Flash proofs finds that F2 accounts for the failure in every case—even though it is the mode least targeted by existing mitigation proposals.

The obvious question this raises is whether it is possible to build a system that doesn’t produce these failures in the first place, as opposed to detecting them after the proof has been written. A prevention-oriented system would need to enforce, during generation, that every load-bearing claim in the proof is either derived from stated premises, grounded in a retrieved and verified source, or explicitly flagged as unverified before the output is returned. The failure modes described here are, I think, a reasonable specification of what such a system would need to prevent.

In fact, as climate models predicted, every drought for the last 30 years in Corpus Christi has exceeded the parameters contemplated in local plans, thanks to fatal delusions deep in the heart of Texas’ methodology: Texas doesn’t plan for droughts to get worse.

What that means for medical AI models is that any patient whose data is used to educate the bot could be exposed, leading to details about their medical history and diagnoses being leaked. In an analysis of seven medical AI datasets consisting of images, ECG records, and general electronic health records, the team determined that individual patients targeted by such attacks can be identified with “near-perfect attack success,” which they explain flies in the face of how such models are evaluated for safety. 

“The fact that MIAs can achieve near-perfect success rates for individual patients is not adequately captured by the standard evaluation protocol, which measures attack success in aggregate across records,” the researchers said. Based on their findings, they conclude, reporting standards for AI privacy audits need to change. 

It gets worse, too: Patients in the dataset are generally easy to identify and, unsurprisingly, those underrepresented in medical AI training data are even easier to finger than those whose data doesn’t stand out.

cross-posted from: https://sopuli.xyz/post/47799383

Labour lost more votes to the left-wing Green Party than to Reform at the local elections last month, polling has shown.

A new study then revealed that over half of former Labour voters who intend to vote for a centre or left-wing party in the next general election cited Israel's genocide in Gaza as a factor in their decision.

The findings indicate the enormous significance the genocide and the UK's cooperation with Israel throughout it have had on Starmer's legacy.

...

...in a recent interview with the News Agents podcast, former health secretary Wes Streeting said that Starmer had accused him of sharing a dossier of evidence of Israeli war crimes provided by British doctors who had been to Gaza for “political purposes”, so that it could be leaked. 

“When I sent that dossier around, the prime minister accused me of sending around a document that was designed to be leaked,” said Streeting.

“I had met British doctors, I had been distressed by what they told me, I had seen serious and substantial allegations of war crimes being committed and I felt this country had a moral and legal responsibility to respond.”

cross-posted from: https://sopuli.xyz/post/47799327

Centrists don't understand that science is already destroyed in the US, the shockwave just takes time to propagate through the system.

Make no mistake, we have no future without science and we have just unplugged science.

The National Science Foundation (NSF) is trimming this year’s budgets for hundreds of its traditional basic science programs by 20% to 30% or more even though its overall budget is down just 3%, Science has learned. NSF has not publicly explained the drastic cuts. But sources within and outside the agency, who did not want to be named, say they suspect the goal is to free up funds for a new $1.5 billion initiative, launched last month, meant to turn NSF-funded discoveries into new products and industries.

...

Program managers would normally rush to tell potential and current grantees about such dramatic changes. But the memo tells program managers to keep their mouths shut. “This information is highly confidential,” it reads. “Please do not communicate anything to PIs [principal investigators].”

Knowledgeable sources within and outside NSF have told Science about comparable cuts in many other units, including 60% for each of the three core research programs within the geosciences directorate. The agency’s biology directorate has been cut by $200 million from its FY 2025 level of roughly $800 million. And some directorates have been hit even harder. NSF is “dissolving” its smallest directorate, which funds social, behavioral, and economic sciences and has made only a handful of awards this fiscal year. The coming months could bring other cuts: For 2 years running, President Donald Trump has proposed slashing NSF’s $1 billion education directorate by nearly three-quarters.

And there is a deeper problem. The actors most capable of destroying the agreement are precisely those least constrained by it. Israel, Hezbollah and the broader network of Iranian-backed militias across the region all sit outside the agreement. They gain little by complying and risk little by defecting because they never signed. A settlement that excludes powerful spoilers has no way to make breaking it hurt.

Centrists don't understand that science is already destroyed in the US, the shockwave just takes time to propagate through the system.

Make no mistake, we have no future without science and we have just unplugged science.

The National Science Foundation (NSF) is trimming this year’s budgets for hundreds of its traditional basic science programs by 20% to 30% or more even though its overall budget is down just 3%, Science has learned. NSF has not publicly explained the drastic cuts. But sources within and outside the agency, who did not want to be named, say they suspect the goal is to free up funds for a new $1.5 billion initiative, launched last month, meant to turn NSF-funded discoveries into new products and industries.

...

Program managers would normally rush to tell potential and current grantees about such dramatic changes. But the memo tells program managers to keep their mouths shut. “This information is highly confidential,” it reads. “Please do not communicate anything to PIs [principal investigators].”

Knowledgeable sources within and outside NSF have told Science about comparable cuts in many other units, including 60% for each of the three core research programs within the geosciences directorate. The agency’s biology directorate has been cut by $200 million from its FY 2025 level of roughly $800 million. And some directorates have been hit even harder. NSF is “dissolving” its smallest directorate, which funds social, behavioral, and economic sciences and has made only a handful of awards this fiscal year. The coming months could bring other cuts: For 2 years running, President Donald Trump has proposed slashing NSF’s $1 billion education directorate by nearly three-quarters.

cross-posted from: https://sopuli.xyz/post/47799327

Centrists don't understand that science is already destroyed in the US, the shockwave just takes time to propagate through the system.

Make no mistake, we have no future without science and we have just unplugged science.

The National Science Foundation (NSF) is trimming this year’s budgets for hundreds of its traditional basic science programs by 20% to 30% or more even though its overall budget is down just 3%, Science has learned. NSF has not publicly explained the drastic cuts. But sources within and outside the agency, who did not want to be named, say they suspect the goal is to free up funds for a new $1.5 billion initiative, launched last month, meant to turn NSF-funded discoveries into new products and industries.

...

Program managers would normally rush to tell potential and current grantees about such dramatic changes. But the memo tells program managers to keep their mouths shut. “This information is highly confidential,” it reads. “Please do not communicate anything to PIs [principal investigators].”

Knowledgeable sources within and outside NSF have told Science about comparable cuts in many other units, including 60% for each of the three core research programs within the geosciences directorate. The agency’s biology directorate has been cut by $200 million from its FY 2025 level of roughly $800 million. And some directorates have been hit even harder. NSF is “dissolving” its smallest directorate, which funds social, behavioral, and economic sciences and has made only a handful of awards this fiscal year. The coming months could bring other cuts: For 2 years running, President Donald Trump has proposed slashing NSF’s $1 billion education directorate by nearly three-quarters.

During our brigade’s recent rotation at the Joint Readiness Training Center, an infantry battalion discovered something essential to success on the modern battlefield. Tasked to screen the approach of a larger air assault, 2nd Battalion, 506th Infantry built a layered web of sensors—company drone operators tied directly to mortar sections, scout teams pushed beyond the forward line of troops, and medium-range reconnaissance drones feeding the targeting cycle. The battalion’s single most effective intelligence asset was that medium-range platform. And the after-action lesson its leaders drew was blunt: The drone and its operators had become such a high-payoff target that the system needed to be flown by a specially trained infantryman who could survive in the close fight—not by a tactical unmanned aircraft system (UAS) pilot trained for employment at higher echelons.

That conclusion, reached independently by a maneuver battalion under combat-realistic conditions, makes the case for a new military occupational specialty in miniature. The proliferation of small UAS, first-person-view (FPV) strike drones, and AI-enabled sensing tools has changed how infantry formations move, hide, communicate, and survive. The question is no longer whether drones matter to the close fight. It is whether the Army’s personnel model is built for the battlefield that now exists. It is not. The infantry needs a dedicated specialty—call it 11R, the drone-enabled infantryman—because drone-enabled warfare has become a persistent, technically demanding function of maneuver that can no longer survive as an informal additional duty without degrading both drone proficiency and infantry fundamentals.