llama.cpp has released version b9694, including binaries for macOS, Linux, Android, Windows, and openEuler. The release supports various architectures and acceleration options such as CUDA, Vulkan, OpenVINO, SYCL, and ROCm. A fix for the Windows x64 OpenVINO release link was also implemented.
A community initiative suggests creating a crowdsourced coding dataset to enable local LLM development. The proposal aims to allow anyone with hardware to contribute data, with more powerful users helping to fine-tune or quantize models, thus reducing reliance on company-released models.
A Reddit user asks the community about their recent projects, noting that while discussions focus on tools, there is little insight into the actual applications or work being done with those tools.
GLM-5.2 demonstrates exceptional long-context coherence and conversational fluency, outperforming Gemini-3.1-Pro on text-only tasks and matching GPT-5.5 in reasoning quality. The model responds factually to sensitive topics like Taiwan and Tiananmen Square, providing detailed historical context without overt censorship, though it adheres to Chinese government content guidelines.
Midjourney has announced a full-body ultrasound CT scanner, calling it the first new whole-body medical imaging modality in 50 years. The prototype, known as the Midjourney Scanner, uses 8,960 transducers across 40 systems in a 70 cm ring to capture data at 17 GB/s, with claimed resolution down to 0.5 mm and a goal of 358,000 ultrasonic elements. The system is currently in Gen 1, with scans taking 20 minutes and no AI used in image generation yet, though future versions aim to integrate AI and reach 50,000 scanners by enabling 1 billion scans monthly.
A Reddit post discusses the potential release of Q.01, noting that precision is no longer a priority. The post highlights a phenomenon referred to as the 'price rising effect' as being significant and unexpected.
Discriminator-Guided RL (DRL) uses a pretrained representation space to train a discriminator that separates real data from model-generated samples. Its logit is used as a reward in KL-regularized RL, aligning model outputs with visual and semantic realism without human preferences. DRL improves FID and semantic FD across models like SiT and JiT, and enhances the Pareto frontier between preference and fidelity.
Essential Subspace Merging (ESM) reduces inter-task interference by focusing on principal directions of activation shifts. ESM++ extends this with dynamic expert selection via prototype-based routing, enabling efficient, training-free multi-task model merging.
Safety Reflection Pretraining inserts short safety reflections into pretraining data to enable self-monitoring in language models. Experiments with 1.7B models on FineWeb-Edu show improved safety accuracy and reduced attack success rates, with MedSafetyWorld demonstrating that the method better prevents unsafe behaviors from being generalized from safe data than data filtering or rewriting.
Stochastic momentum methods like HB and ASGD show distinct batch-size tradeoffs in compute efficiency and serial runtime. HB maintains SGD-level compute efficiency over a batch-size window up to a factor \sqrt{\kappa} larger than SGD's critical batch size, while ASGD improves small-batch efficiency for rapidly decaying spectra but sacrifices it for larger batches in exchange for reduced serial runtime.
AGDN introduces a graph neural network framework that addresses topological priors and connectivity loss in TSP. It uses a MixScore transition matrix and anisotropic diffusion to enable efficient information exchange, outperforming existing methods across diverse problem sizes and distributions while maintaining competitive computation time. The implementation is available on GitHub.
A new metric, Cross-dataset AUC (Cross-AUC), addresses limitations of traditional AUC evaluations by averaging per-domain AUCs and incorporating prediction polarization via Wasserstein Distance. It better reflects real-world performance under domain shifts and provides interpretable insights into detector degradation.
A new automated system addresses extreme class imbalance and asymmetric label noise in Autonomous Emergency Braking data. It uses targeted data augmentation and noise suppression to identify rare delayed and false triggers with 80% improved recall and 50% reduced manual annotation effort, enabling continuous self-improvement in on-vehicle AEB optimization.
A new decision-focused RL framework jointly trains a forecaster and charging controller to handle unknown EV departure times. By aligning forecast accuracy with downstream decision quality, the method achieves up to 14% higher total reward and a 55% reduction in unsupplied energy compared to standard RL approaches.
A new theory models how semantic paraphrases can fool financial sentiment classifiers by analyzing the worst-case displacement of target model representations. The attackability index λ*(x) is derived from the largest generalised eigenvalue of a matrix pencil (A,B), offering closed-form predictions and robustness certificates for affine readouts. The framework connects continuous perturbation theory to discrete paraphrase search, with empirical validation on real financial text classifiers.
XGBoost-Forget enables efficient machine unlearning for XGBoost models on tabular network intrusion datasets. It maintains model performance while achieving faster unlearning compared to full retraining, addressing a gap in unlearning research for tabular data in network intrusion detection.
MAST, a mechanism-guided unlearning method, achieves targeted forgetting of RLVR-induced reasoning with minimal collateral damage. On Qwen2.5-Math-1.5B and Qwen3-1.7B-Base, it significantly reduces MATH performance (45/150 to 37/15-0) while preserving GSM8K accuracy by +0.8 points and maintaining MATH retention at -0.5 points. Results hold across different seeds, objectives, and models, showing superior stability over full-parameter unlearning.
A human-in-the-loop Bayesian optimization framework extends Pareto Front Guided Sampling by incorporating probabilistic constraint satisfaction and input robustness as explicit objectives. It enables domain experts to iteratively refine selection criteria through interactive dashboard projections of trade-offs between performance, uncertainty, and feasibility in CHO cell culture optimization.
STARE addresses policy entropy collapse in GRPO-based reinforcement learning by identifying entropy-critical token subsets via surprisal quantiles and reweighting their advantages. It maintains stable policy entropy across model scales and tasks, outperforming DAPO and other baselines by 4%-8% on AIME24 and AIME25, with consistent exploration-exploitation balance.
TxBench-PP is a verifiable benchmark for small-molecule preclinical pharmacology, testing AI agents' ability to derive accurate conclusions from real-world assay data. Across 16 model-harness configurations, no system reliably made correct preclinical pharmacology decisions, with the best performance at 59.3% (Claude Opus 4.8 / Pi) and 55.3% (GPT-5.5 / Pi) of endpoint attempts.