The Archive

The Trump administration, apprehensive of potential security issues, has reportedly asked OpenAI to stagger the release of its next big-ticket model, GPT-5.6. The Information reported that OpenAI CEO Sam Altman told employees Wednesday in a company Q&A that it would release GPT-5.6 in limited preview form - granting access only to a small group of enterprise customers - in compliance with a request from the federal government. During that preview period, the Trump administration itself would reportedly approve access for customers on a case-by-case basis. It's a more favorable deal than the T...

Agent testing startup Patronus AI, founded by former Meta AI researchers, is experienced nearly insatiable demand, its investor says.

User-built meta-tracker aggregating 17+ status checkers for Fable 5 access restoration.

Notion is "going all in on using agents to run your inbox."

It took 20 years, but the Finance app arrives just in time to be packed full of AI.

Alibaba allegedly used 25,000 accounts to mine Claude over 28.8 million exchanges.

DanceOPD uses on-policy distillation to unify text-to-image, local and global editing in flow-matching models by resolving capability conflicts.

RiVER framework enables RL training of LLMs on optimization tasks without ground-truth answers using continuous execution feedback.

Autoregressive Boltzmann Generators improve molecular sampling efficiency beyond normalizing flows by removing invertibility constraints.

Study quantifies alignment between sequence probability and correctness across LLM decoding methods, models, and benchmarks.

Error-Conditioned Neural Solvers use learned correction instead of gradient descent to improve PDE solutions and extrapolation.

Modular open-source NLP pipeline extracts political-elite networks from multilingual text using joint entity-relation extraction.

Study analyzes BEACON method for low-resource entity matching under varying data constraints and domain awareness.

Language-based digital twin framework using LLMs detects Mild Cognitive Impairment via stylometric analysis and conversation patterns.

PEEU method trains open-source MLLMs for GUI task planning via autonomous environment exploration and hindsight experience replay.

MMBench2 dataset and study show world model hallucinations concentrate in low-coverage state-action regions and are preventable.

Despite ChatGPT's commanding market lead, consumers who pay for AI have been increasingly choosing Anthropic's Claude, data shows.

Sparse autoencoders with sparsity regularizers improve interpretability of vision model activations beyond top-k architectural constraints.

German Central Bank applies LLM-based information extraction to verify securities collateral eligibility from prospectuses, replacing rigid NER.

Gradient equilibrium in online optimization is algorithmically equivalent to Blackwell approachability, clarifying its position in online learning.

Taxonomy of indirect linguistic encoding (algospeak, euphemisms) mechanisms enables LLM detection of camouflaged social media expressions.

Context-aware translation cascades improve multilingual LLM reasoning by retaining original query and reasoning traces across translation stages.

Transfer learning framework combines lightweight physics simulations and autoencoders for structural damage diagnosis with limited experimental data.

Analysis of 15,000 reviews of AI healthcare chatbots identifies reliability, UX, and billing breakdowns; privacy concerns linked to lowest satisfaction.

datasette-export-database 0.3a2 released with dependency constraint fix for Datasette 1.0a27 compatibility.

Fast algorithm learns truncated high-dimensional Gaussians with optimal O(d²/ε²) sample complexity, improving on prior FOCS'24 results.

Analog Interaction Systems framework enables generative modeling on analog hardware (coupled oscillators, Ising machines) with physics-constrained dynamics.

RLAIF framework generates portable job search queries abstracted from seeker identifiers; addresses reward hacking in LLM-as-judge optimization.

Study identifies co-failure ceiling limiting accuracy gains in multi-model LLM routing/voting systems across 67 frontier models; proposes beta metric beyond pairwise correlation.

Empirical study of prompt injection attacks on LLM-based résumé screening; shows effectiveness collapses as injection adoption increases.