How Third Moment Research runs 10,000 agents for 36+ hours
Third Moment Research deploys 10,000 agents for 36+ hour runs using GPT-6 Astra, Sol, Codex, and Daybreak for cybersecurity tasks with persistent improvement.
Search the full wire by company, model, lab, or keyword. Every story we have ever aggregated.
Third Moment Research deploys 10,000 agents for 36+ hour runs using GPT-6 Astra, Sol, Codex, and Daybreak for cybersecurity tasks with persistent improvement.
3D foundation model approach to whole-body PET/CT image registration using learned representations instead of raw intensities.
SolBehave benchmarks behavioral correctness of LLM-generated Solidity contracts via differential fuzzing and transaction replay across 250 Ethereum contracts.
Varest compresses image tokens via semantic variable-resolution tokenization and learned relevance pruning, maintaining visual quality across three backbones.
ProAS resolves credit assignment in LLM agent RL via progress-aware advantage shaping, improving over uniform critic-free methods without adding critic overhead.
Contextual entrainment in LLMs involves shared interventions across relations, not isolated relation-specific heads, affecting robustness to irrelevant context.
Gemini 4 Argon announced as high-capability model targeting complex problem-solving across domains.
We’re giving developers, enterprises, scientists, and consumers access to Gemini 4 Argon, our latest frontier model.
Solar geoengineering simulation shows what happens when geopolitics takes over.
PerturbEnergy uses conditional energy priors to isolate perturbation-specific signals in cellular response prediction for drug discovery.
Today, NVIDIA is making the full NVIDIA Run:ai platform available as open source software, giving customers, developers, and ecosystem partners access to... Today, NVIDIA is making the full NVIDIA Run:ai platform available as open source software, giving customers, developers…
Interactive 4D volumetric liquid forecasting task predicts spatiotemporal fluid dynamics conditioned on solid object motion for robotics and digital twin applications.
Active Probing Benchmark evaluates LLM multi-turn reasoning via generalized puzzle-solving with partial information and concept identification tasks.
Regional photovoltaic generation field forecasting improves ultra-short-term solar prediction by modeling spatial fleet dynamics instead of discrete time series.
Paper identifies failure modes in multimodal entity alignment for knowledge graphs caused by incomplete modalities and structural asymmetries during fusion.
Marvo self-evolves visual agents by using a challenger-solver pair to learn tool invocation from unlabeled images with margin reward feedback.
Tilted temperature decoding rule derived as closed-form solution to pass@k maximization, explaining why higher temperatures benefit larger sample budgets in LLMs.
Trace Evolution Bench protocol isolates genuine agent learning from frozen rescoring effects, scoring self-improving agents on coding tasks with hidden unit tests.
PATTERN benchmark leaks patterns via training labels; simple lookup reaches 99.2% accuracy, revealing fundamental synthetic benchmark flaw.
After a recent spate of high-profile incidents in which AI agents escaped containment, Anthropic is cutting off internet access for all internal evaluations. In a report Friday, the company detailed "unintended model actions," including submitting a false tip regarding an…
Kohli (DeepMind) & Candido (Biohub) discuss limitations of AlphaFold and scaling approach to protein folding beyond structure prediction.
An Anthropic AI model provided false information about an unsolved homicide to a Philadelphia Police Department (PPD) tipline, according to a report from 6abc. In a statement released on Friday, the PPD said the AI model sent the tip through PhillyUnsolvedMurders.com on July…
Gaussian process solvers for linear PDEs with inequality constraints (nonnegativity, monotonicity) via Borel measure conditioning.
Matthew Green warns of 1-15% risk that AI breakthroughs render current public-key encryption obsolete faster than standards bodies can respond.
LISCO learns primal-dual solver steps via neural networks to reduce real-time cost of nonlinear MPC for constrained optimization.
CNN-Transformer with sigmoid label function for direction-of-arrival estimation; signal processing application.
PlacePulse creates transferable geospatial embeddings from structural demographic and business data at fine spatial resolution for foundation models.
BLOKS accelerates kernel ridge regression via sketched block preconditioning to enable scalability without excessive memory overhead.
PDE-PFN applies prior-data fitted networks to PDE solving via in-context learning, using self/cross-attention for posterior predictive inference.
Asana achieved 76x cost reduction and 5x speed improvement in browser agents using GPT-6 Astra with Codex.