'A win is a win' says athlete who didn't let a little diarrhea keep her from finishing in first place Kansas City Chiefs' roster concerns linger as they host the Denver Broncos on Monday Night Football Florida State fires AD Michael Alford, as Mike Norvell's seat reaches a boiling point in Tallahassee Madison Beer's bid to become the NFL's next Taylor Swift hits road bump, Browns are a mess & a Daejon Love fan WWE broadcaster believes'the pieces are all in place' for an upset as Penta challenges Roman Reigns for title Denny Hamlin says forward-facing sonar in professional bass fishing has gone'a little overboard' NASCAR driver says quiet part out loud about Sophie Cunningham, the WNBA looks even worse & car goes airborne! Rams' Sean McVay reveals wife's reality check after mulling media jump following losing season in 2022 Caitlin Clark says she and Angel Reese have'always been friends' after winning FIBA Women's World Cup Stefon Diggs guilty of NFL's dumbest move of Week 1 and it may have decided Commanders loss to Eagles John Harbaugh's Giants on full display in bruising win over Cowboys to kick off NFL season'Tradwife' vs. 'girlboss' debate sparks clash over modern womanhood Americans are right to be scared over'God in a box,' AI CEO says This is causing everybody to rethink what is going on in AI: R'Ray' Wang The WWE star praised Undertaker's intensity and hands-on approach as a booker amid Triplemania 34 The Undertaker ruled the WWE ring for decades as he developed his Dead Man persona and put the fear into any of his opponents that would step up to him. Those days are a thing of the past and The Undertaker has since switched his role from fearsome in-ring competitor to one of the creatives behind the AAA promotion in Mexico. Undertaker looks on during NXT at 713 Music Hall in Houston, Texas, on July 22, 2025. The WWE legend has been among the minds behind AAA's rise to prominence on the pro wrestling scene.

'Team Mom' Alison Lee completes largest comeback in Solheim Cup history for Team USA to cap off unbeaten week Texas coach Steve Sarkisian's awkward ESPN interview divides internet during team's win over Ohio State'RedZone' returns for 18th season and, despite 15-second commercials, could still be on in the White House Dan Dakich says women might be the reason for Nike's downfall: 'They p----d them off' College football's chaos era is here: Texas rallies, Oregon falls and Lane Kiffin eyes return to Ole Miss America's women's basketball epic reaches its climax with national unity on the line in FIBA World Cup final Oklahoma State stuns Oregon, deals early season blow to Ducks' national title hopes Trump slams Europe's left-wing policies as Canada seeks EU ties White House advisor calls on tech leaders to'stop pretending' over AI threat Canada explores'associate member' status with EU amid US trade tensions Trespasser stopped by security at Kamala Harris' Malibu property Dominik Mysterio will essentially pull double duty at Triplemania 34 . The AAA mega champion was in Omos' corner when the 7-foot-3 giant defeated Rey Mysterio on Night 1 in Las Vegas on Friday night at the MGM Grand Garden Arena. On Night 2, Mysterio will defend the title against El Grande Americano. Dominik Mysterio and JD McDonagh make their entrance during Monday Night Raw at Rocket Arena in Cleveland, Ohio, on Aug. 31, 2026. Mysterio has morphed into one of the top talents in pro wrestling whether he's seen on WWE or AAA programming.

Large language models (LLMs) are increasingly deployed in high-stakes domains, where free-text explanations such as chain-of-thought and post-hoc rationales are used to justify model outputs. Yet it remains unclear whether these explanations are sufficient, i.e., if they contain enough information to explain the model's output-generating process. We generalize classical sufficiency from feature attributions to arbitrary explanations and prove that explanation sufficiency can change depending on the input distribution, which must be explicitly defined for LLM explanations. We propose using the LLM itself to generate alternative inputs conditioned on an explanation, capturing its beliefs about possible inputs. We formalize self-consistent sufficiency as a goal for free-text explanations and introduce an information-theoretic metric, SCSuff, that enables evaluation of free-text explanations without relying on predefined biases or shortcuts. Our experiments show that SCSuff agrees with targeted perturbation tests where applicable and demonstrate that explanation sufficiency can vary with the input distribution. We find LLM explanations are generally insufficient and weakly correlated with model size, accuracy, or output entropy. Analysis of final-token hidden states shows that top and bottom SCSuff scores can be predicted from internal representations, suggesting that SCSuff can guide detection and improvement of sufficient LLM explanations. The code for this paper is available at https://github.com/rajesh-lab/self-consistent-sufficiency .

This material may not be published, broadcast, rewritten, or redistributed. Quotes displayed in real-time or delayed by at least 15 minutes. Market data provided by Factset . Powered and implemented by FactSet Digital Solutions . Mutual Fund and ETF data provided by LSEG . China's brain chip breakthrough raises big questions Should you change your phone number after a hack? McDonald's AI drive-thru may take your next order The Father's Day gift that protects your dad from scammers New Air Force One'flying palace' gifted by Qatar unveiled for President Trump Kevin O'Leary warns U.S. must accelerate data center growth to keep pace with China in AI race Americans need protection against'warrantless surveillance': Rep Chip Roy Spencer Pratt's use of AI to boost campaign sparks debate China approves world's first commercial brain chip A suspicious AAA-themed email warns about a supposed car safety rule.

The score-based query attacks (SQAs) pose practical threats to deep neural networks by crafting adversarial perturbations within dozens of queries, only using the model's output scores. Nonetheless, we note that if the loss trend of the outputs is slightly perturbed, SQAs could be easily misled and thereby become much less effective. Following this idea, we propose a novel defense, namely Adversarial Attack on Attackers (AAA), to confound SQAs towards incorrect attack directions by slightly modifying the output logits. In this way, (1) SQAs are prevented regardless of the model's worst-case robustness; (2) the original model predictions are hardly changed, i.e., no degradation on clean accuracy; (3) the calibration of confidence scores can be improved simultaneously. Extensive experiments are provided to verify the above advantages. For example, by setting $\ell_\infty=8/255$ on CIFAR-10, our proposed AAA helps WideResNet-28 secure 80.59% accuracy under Square attack (2500 queries), while the best prior defense (i.e., adversarial training) only attains 67.44%. Since AAA attacks SQA's general greedy strategy, such advantages of AAA over 8 defenses can be consistently observed on 8 CIFAR-10/ImageNet models under 6 SQAs, using different attack targets, bounds, norms, losses, and strategies.

Following this idea, we propose a novel defense, namely Adversarial Attack on Attackers (AAA), to confound SQAs towards incorrect attack directions by slightly modifying the output logits.

How effectively can LLM-based AI assistants utilize their memory (context) to perform various tasks? Traditional data benchmarks, which are often manually crafted, suffer from several limitations: they are static, susceptible to overfitting, difficult to interpret, and lack actionable insights--failing to pinpoint the specific capabilities a model lacks when it does not pass a test. In this paper, we present a framework for automatically generating a comprehensive set of tests to evaluate models' abilities to use their memory effectively. Our framework extends the range of capability tests beyond the commonly explored (passkey, key-value, needle in the haystack) search, a dominant focus in the literature. Specifically, we evaluate models on atomic tasks such as searching, recalling, editing, matching, comparing information in context memory, and performing basic operations when inputs are structured into distinct blocks, simulating real-world data. Additionally, we design composite tests to investigate the models' ability to maintain state while operating on memory. Our benchmark enables an interpretable, detailed assessment of memory capabilities of LLMs.