Reasoning

TL;DR for operators A model generates an incorrect answer. An automated verifier can reject it but cannot explain the mistake. ReTeach1 shows how that failure can become training material without a worked solution or a larger external teacher. The model’s teacher revisits unsuccessful attempts, reflects on possible errors, and retries. But the successful retry is not simply copied into the student’s training targets. Instead, the teacher uses its additional reasoning context to produce better-informed token predictions along the student’s original response. The student learns from those predictions without needing that context itself. On SciKnowEval-Physics, explicit reflection achieves 64.24% accuracy, versus 62.91% with retries but no reflection (Section 4.3, Figure 4). ...

TL;DR for operators An autonomous purchasing agent may be choosing between two reasonable products while the interface quietly changes the decision environment: one item is preselected, or one is presented as socially popular. A randomized experiment with 3,600 GUI agents and 21,600 product choices found that both interventions strongly shifted what agents selected. More strikingly, increasing the agents’ reasoning setting pushed the two effects in opposite directions. Within the default condition, high reasoning roughly halved the odds of selecting the nudged product. Within the social-influence condition, high reasoning increased those odds by about 71%.1 ...

TL;DR for operators Kim and colleagues’ study of pause-token fine-tuning1 points to a training rule that is easy to miss if pause tokens are treated mainly as extra thinking time. At an equal pause-token budget, putting pauses at semantic boundaries is the only tested placement that consistently improves both math and code averages over ordinary supervised fine-tuning. The stronger variant also masks the loss on the pause tokens themselves. ...

TL;DR for operators UrduBench1 tests 23 open and open-weight models on 2,390 held-out Urdu questions spanning arithmetic reasoning, formal mathematics, commonsense, and knowledge tasks. The ranking gives little support to selecting an Urdu model from parameter count or a “reasoning” label alone: Gemma-3-12B-it leads the reported aggregate at 59.4%, while the larger reasoning-oriented DeepSeek-R1-Distill-Qwen-14B scores 44.9%. ...

TL;DR for operators When an LLM has to plan, call a tool, inspect the result, remember what happened, and decide what to do next, the system has changed in a more fundamental way than “adding more reasoning steps.” Agentic Reasoning for Large Language Models frames that change as a move from mostly static generation toward an interactive reasoning-and-control loop.1 ...

TL;DR for operators When one model answer can trigger a consequential downstream step, spending more inference compute before trusting that answer can help—but the form of that spending matters. In the reported experiment, generating several reasoning attempts and aggregating the answer that repeatedly emerged increased verified acceptance from 56.2% to 64.9%. Asking the same model to critique and revise itself produced a smaller increase, from 47.2% to 50.6%. Adding a second model did not improve the reported acceptance rate: 47.4% of outputs survived cross-model verification versus a 48.7% single-model baseline. ...

TL;DR for operators When a model is midway through a difficult task, the expensive choice is often whether to leave it alone, verify it, or spend more inference on correction. Doing that for every request wastes compute and can damage answers that were already on track. Lihao Sun and colleagues show that the model’s internal state changes systematically as reasoning progresses, and that late-stage changes contain a useful warning signal for eventual failure.1 Their trajectory-based correctness features reach a seed-averaged best-layer ROC-AUC of about 0.85, versus 0.765 for the strongest reported LogitLens baseline and 0.649 for reasoning step count alone. ...

TL;DR for operators A correct multimodal answer with citations can still have a weak evidence trail. In human evaluation on Video-MMMU, Gemini-3-Pro reached 86% question-answering accuracy but only 16.3 on MuRGAt-Score, meaning answer performance and fact-level grounding ranked very differently. MuRGAt evaluates two requirements separately. Coverage asks whether verifiable claims receive citations. Attribution asks whether the cited frames, audio, figures, or temporal segments are actually relevant to and sufficient for the factual claims they accompany. The combined score rewards systems only when both conditions hold. ...

TL;DR for operators Synthetic data should not be budgeted as a cheaper substitute for human examples. It should be treated as infrastructure for producing controlled training experience. The operational sequence is: generate tasks that can actually be executed and scored; verify and repair them before spending compute on trajectories; choose a training objective that reinforces the capability you want rather than merely reproducing successful-looking behavior; and test the resulting model outside the environment in which that experience was generated. ...

TL;DR for operators A quality-control component can correctly identify a defect and still add little value if the next stage ignores the correction. That distinction matters for AI workflows built around critics, reviewers, validators, or approval agents: reviewer accuracy measures whether the warning is right, not whether the warning changes what the system ultimately does. ...