The Archive

Statistical analysis of 13 neural audio codec token sequences across architectures and noise conditions using Zipf/Heaps metrics.

Normalized Low-Rank Adaptation (NoRA) stabilizes LoRA training by normalizing down-projection matrices, improving parameter-efficient fine-tuning.

Study evaluates when learned perceptual predictors align with human listeners in codec-based speech models using GRPO with CER constraints.

Probes autonomous driving benchmark scores by removing camera input, measuring how much performance relies on memory vs. dynamic scene understanding.

Framework learns latent representations of PDEs by embedding scientific inductive bias into training distributions for hypothesis discovery.

MNIST-PRO benchmark isolates agentic perception by converting digit recognition into sequential glimpse-based search with memory constraints.

Audit of three AI clinical scribes on 142 consultations found 31.3% of notes contain verified errors, concentrated in allergy and medication data.

The three-year-old startup says it reached $15 million in ARR and profitability before raising its latest $15 million round.

A perception stack is shaped by the vehicle that carries it. Move the same software to a new carline—for example, from an SUV to a sedan or another vehicle... A perception stack is shaped by the vehicle that carries it. Move the same software to a new carline—for example, from an SUV to a sedan or another vehicle variant in the portfolio—and its perception of the world changes. The sensor placement, calibration, fields of view, occlusions, body geometry, timing, and coverage all shift. A traffic light may appear in a different part of the frame. Source

LLM judges fail to detect omissions in clinical notes; benchmark with 500 single-error pairs shows judges verify presence, not absence.

Framework addresses evidence-bounded mental health reasoning from heterogeneous speech protocols, preventing hallucination across different clinical tasks.

TSPFN adapts TabPFN foundation model architecture for physiological time-series classification with temporal dependencies in low-data regimes.

LiFT uses language-informed Flow Matching with agent-guided SMILES generation for 3D molecular design balancing affinity and chemical validity.

Clinical NLP framework uses automatic language analysis to augment psychiatrist judgments of patient experience in therapeutic conversations.

Study finds KV-cache quantization in RAG systems reduces faithfulness to retrieved evidence even when accuracy holds on Qwen2.5-7B.

Knowledge-aligned SFT framework constrains fine-tuning targets to base model's parametric knowledge to reduce hallucinations via Evidence Rewrite and Recall variants.

Stiefel-constrained Riemannian optimization enables parameter-efficient rotational activation steering to control LLM refusal behavior at inference.

New benchmark and synthetic training pipeline improve LLMs' self-modeling—ability to predict own behavior on counterfactual prompt edits.

Sparse competition-based training induces modular specialization in neural networks while maintaining baseline accuracy.

CastClaw: human-in-the-loop LLM agent system for industrial time-series forecasting with domain-specific constraints and uncertainty communication.

MR-JEPA: self-supervised video foundation model for cardiac MRI pretraining on multi-sequence data from 10,505 patients using 3D spatiotemporal masking.

CoJEPA combines JEPA and contrastive learning for music representation learning, balancing local task performance with global representation strength.

Debian voted to allow developers to use AI tools in their contributions to the Linux distribution's "development, maintenance, [and] documentation." The new policy on AI acknowledges that "responsible" use of AI can improve developers' productivity, and goes on to say, "generative AI is neither exempt from nor subject to special rules beyond the standards already expected of Debian contributors." The project's voting developers considered several other proposals on how to handle AI, including some that would have banned contributions made using AI tools. As It's FOSS reports, some users and c...

Speculative philosophical narrative about AI agents forming emergent social structures by 2026, invoking social contract theory.

Nvidia invests $3.5 billion into Taiwanese chipmaker MediaTek. The deal shows how Nvidia plans to stay essential to AI infrastructure as Big Tech begins to build its own AI chips.

Anthropic outlines updates to alignment and security research, internal testing processes, and risk mitigation strategies for AI systems.

Today, I’m talking with New York Governor Kathy Hochul, and I’ll just warn you — this episode moves really fast. It’s an election year, after all, with a shocking amount of tech policy at stake, and Governor Hochul has taken strong positions on almost every major tech issue there is. For example, Meta just reached a settlement with dozens of states, including New York, which will restrict how teens use platforms like Instagram in very specific ways. Governor Hochul is a strong supporter of those restrictions and more, as you’ll hear. But those come with a cost — widespread age verification me...

Explosive growth comes with a new regulatory burden in the European Union.

Import AI newsletter covers Hugging Face concerns, space mining applications, and Five Eyes AI governance positions.

OpenAI will soon be held accountable for mitigating risks related to ChatGPT's impact on minors, user mental health, and the spread of illegal content in the European Union. That's because ChatGPT is now considered a Very Large Online Search Engine under the EU's Digital Services Act, a set of laws regulating major online services and platforms. Along with ChatGPT, the European Commission also announced that Reddit and Roblox are considered Very Large Online Platforms, subjecting them to the same rules. The DSA restricts platforms from targeting ads to minors or using a person's sexual orient...