The Archive

MoPET applies mixture-of-experts to medical image classification, routing tasks across specialized adapters to prevent negative transfer in multi-domain PEFT.

Theoretical analysis of bandit algorithms under heavy-tailed reward distributions without knowledge of tail exponent; resolves COLT open problem.

TFGformer applies retrieval-augmented generation to multivariate time series forecasting via time-frequency graph learning for IoT sensor data.

Simulation study comparing confidence interval coverage across ML algorithms for nuisance parameter estimation in Double Machine Learning treatment-effect inference.

datasette-agent 0.4a0 adds browser task execution, letting agent plugins run JavaScript directly in user browsers.

QR-STT method crafts thermal adversarial patterns to steer infrared vision-language models via optimized thermal states; demonstrates robustness risk.

Framework for end-to-end fairness optimization in ML prediction-to-decision pipelines using group-based alpha-fairness measures.

Analytic memory abstraction for multimodal agents enabling filtering, aggregation, and temporal reasoning over accumulated observations.

When the phrase "OpenAI hacked Hugging Face" has more or less entered mainstream culture, you know we have an AI problem. This week, we learned more about exactly how OpenAI's agent broke out of a sandbox and autonomously traversed the web, including a bunch of other supposedly secure web services, all in the name of cheating on a benchmark tests. The fact that this hack happened is a problem. So is the fact that it took a while for anyone to notice. And the fact that it seems no one is willing or able to do much to stop it. (And lest you think it's just an OpenAI problem, since we recorded t...

The AI chatbot was more effective at creating “exploitable trust” than the humans.

After years of pushing full speed ahead on AI, OpenAI CEO Sam Altman says maybe it’s time for the AI industry to “pace” itself. The comments came just days after one of OpenAI’s own models broke out of its test environment and got tangled up in a breach at Hugging Face — though as Equity’s hosts point out, sloppy security seems to have […]

Decoupled evaluation paradigm isolating memory retrieval vs. utilization failures in LLM personalization and preference incorporation.

ModelEquivBench certifying benchmark for multi-relational evaluation of LLM-generated optimization models with semantic profiling.

AgenticRepair framework augments agentic AI program repair with security-specific context engineering for vulnerability patching.

ENTINEX method for sparse-reward RL exploration using entropic information to incentivize boundary-aware state discovery.

Anthropic just realized several of its Claude AI models hacked into the systems of three different organizations during testing, acting on their own and without the company noticing. The revelation comes days after rival OpenAI said one of its own models had breached developer platform Hugging Face, adding to growing unease over whether frontier AI labs are doing enough to control the increasingly capable systems they are building. In a blog post describing the incidents, Anthropic said Claude gained unauthorized access to the systems during cybersecurity evaluations. All of the attacks happe...

Survey of 257 papers on validation and assurance for multi-step agentic AI systems spanning evaluation, runtime monitoring, and regulation.

Hypothetical Prompt Embeddings (HyPE) reduces computational overhead in RAG by pre-training query-document alignment without runtime generation.

ALIVE auditable control layer for budgeted multi-source learning with randomized warnings and capacity-feasible exclusion certificates.

OnlineCache enables adaptive, error-correcting caching policies for diffusion model inference tuned per-prompt and per-timestep.

Empirical study of quantization trade-offs (latency, throughput, quality) for EuroLLM and Hy-MT2 translation models on A10 GPU.

Medical imaging paper on synthesis of gadolinium-free dynamic MRI using latent transport—outside AI infrastructure scope.

Benchmark comparing 10 LLMs and 6 prompting strategies for automated code generation in fluid system simulation (WNTR, Modelica).

HDFS log anomaly detection paper—systems observability research, not frontier AI.

Black-box LLM inversion via previous-token prediction (PTP) for near-exact prompt reconstruction without weight/logit access.

Zero-Mem: Zero-token memory operations for LLM agents using encoder computation instead of LLM calls to reduce latency and token costs.

Finite-horizon regret analysis for regularized greedy multi-armed bandit algorithms—theoretical ML, limited frontier AI relevance.

Graph domain adaptation method modeling propagation resolution shift for class-discriminative knowledge transfer across graph distributions.

Autoregressive speech generation using low-frame-rate high-dimensional continuous tokens to balance stability and reconstruction fidelity.

Cross-lingual transfer study across five Turkic languages using mT5, charting strongest source-target pairs for low-resource MT.