AI Research

True Positive Weekly #182

The most important artificial intelligence and machine learning news and articles

Jev Your Engines: Seven Open Decision Models vs. a Generative LLM

A hands-on Jev repo comparison of routing accuracy, latency, and output reliability The post Jev Your Engines: Seven Open Decision Models vs. a Generative LLM appeared first on Towards Data Science.

What OpenAI’s 719 Math Papers Mean for Data Science

Matching messy text, choosing representative examples, and generating comparison data. The post What OpenAI’s 719 Math Papers Mean for Data Science appeared first on Towards Data Science.

How Many Guesses Does It Take to Crack an AI's "Random" Number?

I read the exact answer distribution from the logits of open models, base and instruction-tuned, and turned it into bits and guesses-to-crack. The post How Many Guesses Does It Take to Crack an AI's "Random" Number?…

I Tested Four Ways to Give an AI Assistant a Memory. The Obvious One Was the Worst Value.

Pasting in the whole chat history cost 20 to 31 times more than looking up notes, and it did no better on the tasks. The post I Tested Four Ways to Give an AI Assistant a Memory. The Obvious One Was the Worst Value.…

😸 Gemini now screens scam texts

An AI Label Can Change What People Believe, Even When the Content Is True

Google's SynthID Detector makes AI-generated media easier to identify, but studies reveal an unexpected consequence: AI labels can reduce the perceived accuracy of truthful information. The challenge is distinguishing…

GPT-6 Passed One Safety Test at 99.99%. Another Told a Different Story.

OpenAI's October GPT-6 safety report reveals a striking contrast: near-perfect resistance to instruction overrides and persistent warning circumvention in simulated work environments. The findings expose a broader…

Grok Voice Says It Resolves 70% of Starlink Calls. What Counts as Resolved?

Grok Voice reportedly resolves 70% of Starlink’s customer calls with no human intervention. Its launch puts the focus on how businesses verify customer outcomes and measure the total cost per resolved case.

The Next AI Cybersecurity Problem Could Run on Cheap TV Boxes

Anthropic ran a 26-billion-parameter model across $55 TV boxes and tested simplified replication workflows. The findings point toward a possible cybersecurity threat from small models that adapt to unfamiliar devices.

From Global to Tail Convergence: Dynamics of $\alpha$-Stable Score-Based Generative Models

arXiv:2610.19773v1 Announce Type: new Abstract: Score-based generative models driven by $\alpha$-stable L{\'e}vy processes have been proposed as an alternative to Brownian diffusion models, with empirical evidence…

Online Tokenization with Optimal Delay in Linear Time

arXiv:2610.19139v1 Announce Type: cross Abstract: Tokenization is an essential preprocessing step in almost all large language models (LLMs). One of the most popular frameworks is BPE tokenization, which consists of…

OScaR: The Occam's Razor for Extreme KV Cache Quantization in LLMs and Beyond

arXiv:2605.19660v3 Announce Type: replace Abstract: The rapid advancement toward long-context reasoning and multi-modal intelligence has made the memory footprint of the Key-Value (KV) cache a dominant memory bottleneck…

Spectral structure learning for clinical time series

arXiv:2502.11680v2 Announce Type: replace Abstract: We develop and evaluate a structure learning algorithm for clinical time series. Clinical time series are multivariate time series observed in multiple patients and…

MCAIR: Monitoring and Correcting Agent Unfaithfulness via Internal Representations

arXiv:2610.19679v1 Announce Type: cross Abstract: Large Language Model (LLM)-based agents have demonstrated remarkable capabilities in completing complex tasks through multi-step reasoning and tool interactions.…

One Task, Many Interfaces: Simple Cross-Interface Consistency Training Improves LLM Agent Robustness

arXiv:2610.20576v1 Announce Type: new Abstract: Post-training binds the competence of an LLM agent to the interface it was trained on. Semantics-preserving perturbations of that interface, such as renaming an action or…

Bridging the Topological Divide: Dual-Branch Graph Distillation with Semantic Priors for Robust Cold-Start Recommendation

arXiv:2610.19578v1 Announce Type: cross Abstract: Graph Collaborative Filtering models have achieved remarkable success in recommendation systems. However, they inherently suffer from a catastrophic representation…

Towards Training and Inference of Graph Neural Networks in Graph Databases

arXiv:2610.19142v1 Announce Type: new Abstract: Graph Neural Networks on large graphs commonly rely on in-memory graph storage with ad hoc data processing pipelines for both training and inference. Graph databases…

Njord: A Probabilistic Graph Neural Network for Ensemble Ocean Forecasting

arXiv:2605.15470v3 Announce Type: replace Abstract: Ocean dynamics are inherently chaotic, yet existing machine learning ocean models produce only deterministic forecasts. We introduce Njord, a probabilistic data-driven…

How Many Ranks Does a Token Need? A Noise-Floor Squelch for Rank-One LoRA Experts

arXiv:2610.19033v1 Announce Type: new Abstract: Low-rank adaptation (LoRA) applies every rank-one component of its update to every token, and mixture-of-experts variants replace this with a learned router that activates…