Jadavpur University Finds Fine-Tuned PEGASUS Beats GPT-3.5 at Writing Research Titles
A fine-tuned PEGASUS model beats GPT-3.5 and LLaMA-3-8B at generating research paper titles from abstracts, with a new social sciences dataset to boot
20 stories
A fine-tuned PEGASUS model beats GPT-3.5 and LLaMA-3-8B at generating research paper titles from abstracts, with a new social sciences dataset to boot
An 86M-parameter model trained on an Indonesian teaching method beats models 100x its size at basic arithmetic, revealing how structured pedagogy shapes neural computation
AutoLab tests 17 frontier models on 36 multi-hour engineering tasks, revealing that persistence beats raw intelligence every time
A new research paper introduces UniCAD, a 1.4M-model benchmark and unified MLLM that handles CAD reconstruction, generation, and question answering in one framework
A new paper proves exactly which entropy corrections Active Inference needs to turn standard variational inference into full EFE-based planning, with a working message-passing scheme to match.
SharedRequest hides your prompts by mixing them into noisy batches, cutting query costs by 5x while outperforming differential privacy baselines by 20%
M³Eval is the first benchmark to test video AI memory using cognitive psychology experiments, exposing deep flaws in today's best models
DAR gives LLMs on-demand access to legal statutes as tools, pushing state-of-the-art on rule-based reasoning — but only for the strongest models
A new benchmark and taxonomy expose how AI text detectors are built on incompatible assumptions that don't reflect how people actually write with AI
Researchers release PWR, a benchmark and dual-model architecture that teaches AI assistants to watch, plan, and recover when users go off-script during real-world tasks
A new survey maps the full stack of LLM agent accountability — from tool calls and memory writes to multi-agent failures — and argues that final-answer accuracy is no longer enough
CHERRL gives researchers a controlled lab to reproduce, measure, and automatically detect reward hacking in LLM-judge-based training pipelines
AdaKoop turns nonlinear time series into a linear problem, enabling real-time streaming forecasts across 71 benchmarks without retraining from scratch
BiasGRPO applies group-relative reinforcement learning to social bias mitigation, outperforming DPO and PPO with a tiny, plug-and-play 0.1B reward model released on Hugging Face.
VRPCoder, an 8B model trained with constraint-aware reinforcement learning, beats Gemini and Claude on vehicle routing code generation by catching bugs that objective-only testing misses
DeliChess is a new dataset of 107 group chess puzzle dialogues that proves deliberation improves collective reasoning — but probing questions don't always help.
A new VAE-based framework generates adversarial scenarios during RL training to tighten probabilistic safety guarantees, making policy verification sharper and more deployable.
OA-CutMix fixes CutMix's broken label logic using SAM segmentation masks, beating 10+ dynamic mixing methods with zero training overhead
LifeSkill teaches LLM agents to bake experience into their weights at test time, beating all retrieval-based and RL baselines by 7 points on a lifelong learning benchmark.
A new benchmark tests whether LLMs can reason like scientists — and finds that the ability to seek disconfirming evidence is the single biggest predictor of success