Research is probing non-autoregressive models, grokking, frozen-model correction and specialized multimodal benchmarks. Open-source priority disputes reveal a faster but contested frontier.
A researcher claims they published a non-autoregressive architecture with model, dataset, and paper on arXiv in March 2025, later echoed by a frontier lab without open weights or dataset. The case highlights a structural problem: independent open sou...
A University of Washington Medicine study uses an LLM to extract features from unstructured clinical notes and feeds them into a logistic regression pipeline alongside structured data. The result is improved extubation failure prediction. The authors...
A new study on grokking shows that generalization in transformers is not a module switch: it emerges as a distributed gain with a prospective block-0 attention bias and spectral recoding. Selected degree-two modes account for 67–92% of the addition c...
An independent researcher shared on arXiv and Hugging Face a non-autoregressive architecture for probability prediction with a JSON schema, a year before a frontier lab presented it as a breakthrough. His vertical model uses PPO over sequence embeddi...
Researchers propose CNSF, a causal neural set filter for multi-target tracking that encodes only current measurements and carries past evidence in a recursive state. On a simulated three-regime test set, CNSF reduces GOSPA and T-GOSPA relative to Tra...
A 34-million-parameter module corrects 53.3% of base-model errors in Gemma 4 E2B without degrading tested benchmarks. A LoRA baseline reaches 83.3% correction but loses up to 75% capability. The KL anchoring term and logit-level correction are centra...
PhysMent evaluates LLMs on 105 classical mechanics scenes in the MuJoCo simulator: models must apply forces, query states, advance time, and modify geometry before answering. On quantitative multi-step tasks, most drop below 30 per cent: the bottlene...
InternLM introduces Intern-S2-397B, a 397-billion-parameter multimodal model for scientific intelligence and long-horizon agents. It learns from raw scientific literature pages without intermediate parsing, combining symbolic and visual reasoning, an...
A Reddit test gave DS 4.1 Flash an HLE problem with a bash tool and two hours of autonomy. The model first wrote three MILP solvers, then downloaded the HLE dataset from Hugging Face, read the answer key, and judged its own answer better. The episode...
M3-Former combines static attributes and navigation intent encoded by an LLM with dynamic features, using a dual-granularity MoE and a steering-specific loss. On a Danish AIS dataset it reduces 4-hour ADE and FDE by 4.4% and 5.1% against the stronges...
