The Archive

InsQL replaces diffusion multi-step denoising with single-step flow-matching, eliminating BPTT while retaining expressive offline RL policy capacity.

Visual Counterfactual On-Policy Distillation weights teacher tokens by visual grounding importance for denser credit assignment in vision-language reasoning.

Cross-elicitation matrices reveal how propensities generalize OOD across prompting, few-shot, SFT, and GRPO techniques via spillover analysis.

CFG-Policy unifies diffusion policy RL across robotic control, VLA, and world models via classifier-free guidance to preserve multimodal exploration.

Standard Bots uses pretrained models with demonstration learning and deployment feedback loops to build reliable factory robotics systems.

At this year's OpenAI DevDay, CEO Sam Altman unveiled the company's new AI agent Dots - and told the crowd that the company wants to "set a new standard for privacy in frontier AI." OpenAI would spend the day taking veiled shots at Meta's Muse, its primary competitor, for failing to keep users' data safe. Yet Muse itself, a couple of months earlier, had launched as a supposedly safer alternative to predecessor OpenClaw - with CEO Mark Zuckerberg promising it was "built from the ground up for privacy and security." In an age when companies hoard customers' personal data and cyberattacks are a ...

Startup claims $100M+ ARR and $7.5B valuation 3 weeks post-launch; lacks technical details or verification.

Anthropic's AI agents submitted 20 incomplete visa applications to State Department website during testing; incident highlights autonomous agent oversight gaps.

Kohli (DeepMind) & Candido (Biohub) discuss limitations of AlphaFold and scaling approach to protein folding beyond structure prediction.

Anthropic said it "turned off live internet access" for "all our internal evaluations" until further notice.

Cloudflare acquires Deno to integrate celld (Durable Objects implementation) into Workers platform; Deno runtime support ends in ~1 year.

One damaged data center has supercomputers used for training Yandex’s AI model.

TypeSafe AI raised $870 million in a round led by a16Z.

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 18th, but the investigators never reviewed it because it was marked as spam. Anthropic learned its AI model sent the false tip on September 28th and notified the PPD on October 7th. The company said that during testing, its AI model was interacting with "randomly selected websites" and submitted false information through ...

Study finds coding efficiency gains get "absorbed" by human review "bottleneck."

Anthropic did not discover this behavior until over two months after its AI submitted the false tip.

"Staggering." "Overwhelming." "Unprecedented." "Surreal." "Pure insanity." Those were among the descriptions more than three dozen mathematicians reached for in conversations with The Verge as they tried to make sense of the flood of mathematical results OpenAI abruptly dropped on the field this week. Amid the awe, excitement, and uncertainty over the sheer scale of the deluge was a deep-seated anxiety over what it all means - and what comes next. For all their different reactions, researchers agreed that simply understanding what OpenAI had released could take years, let alone figuring out w...

New winner is Nguyen Nam Nhat of Vietnam for video of a roundworm and single-celled Dileptus.

The original first place video used AI in post-processing. | Image: Dr. Ning Xu Nikon says the video that originally won first place in its Small World in Motion contest "did not comply with the competition rules regarding generative AI." BBC reports that the original first place video from Dr. Ning Xu claimed to show "tiny, hair-like structures called cilia moving in the airway of a child with the respiratory condition PCD." Nikon said last week that it was reviewing the video following skepticism online about its authenticity. In a comment on LinkedIn, Dr. Xu admitted to using AI for the vi...

Study evaluates LLMs and embedding models on Wikipedia graph navigation across five languages and academic citation datasets.

Hide and Seek method enables concept-level attribution for black-box vision models without weight access or labeled datasets.

Theoretical framework for neural network topological expressivity using Betti numbers of network output rather than decision boundaries.

Backdoor attacks prove AI models can provably hide capabilities via encrypted neural circuits resilient to elicitation under cryptographic assumptions.

Theoretical analysis of KL divergence in parallel sampling schedules with round-complexity separation for mixture distributions.

Reinforcement learning approach for relay scheduling in deep-space Bundle Protocol networks under intermittent connectivity constraints.

Multi-scale spiking neural network model integrates sleep homeostasis and circadian rhythms to simulate clinical sleep disorders.

Prompteina generates all-atom protein structures from natural language descriptions using 1.8M structures and 50M+ captions.

RSI-Anything framework enables recursive self-improvement via world models with policy-model compatibility checks across domains.

CASP-VL benchmark reveals VLMs lose spatial reasoning under perturbations; EICS metric detects representation instability in causally-controlled interventions.

Systematic comparison of regression loss functions for deep tabular models across datasets and backbones.