Research
-
Research
Fabricated dashboards can make LLM agents act on unknowable outcomes
A reproducible arXiv study finds that fabricated authoritative-looking panels can push some LLM agents to act on unknowable outcomes despite recognizing the uncertainty.
-
Research
openJiuwen turns the coding-agent harness into a measurable systems layer
openJiuwen proposes a composable, runtime-adaptive coding-agent harness and shows why model-matched benchmark audits matter for agent engineering.
-
Research
Self-evolving AI agents can turn poisoned experience into persistent skills
New research shows self-evolving agents can store poisoned interactions as reusable skills, making skill provenance, validation and rollback part of the security boundary.
-
Research
AdaptRubric gives GUI agents task-specific rules for judging success
AdaptRubric builds task-adaptive rubrics for GUI-agent evaluation, with reported gains in reward-model F1 and reinforcement-learning task success.
-
Research
Recuris improves long-horizon agents by evolving memory, not model weights
Recuris keeps the base LLM frozen while evolving working and experiential memory, with reported gains across long-horizon agent benchmarks.
-
Open Source
FreeToken runs frontier-scale MoE models on consumer GPUs
FreeToken uses bandwidth-adaptive CPU-GPU execution and semantic caching to serve 35B to 753B MoE models on consumer and workstation hardware.
-
Research
Inherent trains 27B Faraday agent to replicate research
Inherent’s 27B Faraday agent uses Replica and coding agents to reproduce scientific results, while benchmark caveats complicate direct model comparisons.
-
New Models
Ant Ling opens six Ling 3.0 checkpoints for custom LLM training
Ant Ling releases six Ling 3.0 base checkpoints across two model sizes and three training stages for continued pretraining and specialized LLM work.
-
Research
OpenAI reports ten advances in mathematics and theoretical computer science
OpenAI has published a collection of ten research advances spanning mathematics and theoretical computer science. The work includes results related to long-standing open problems in areas such as geometry, cryptography a…
