MEV: A Multi-Event Video Dataset for Long-Take Generation
Current datasets for text-to-video generation are largely limited to short clips depicting a single event or coarse-grained…
News
Research & Science
Papers, breakthroughs and what the labs found this week.
585 stories · Every headline links to the original publisher. HapnGo curates and groups; we do not copy articles.
The Indian Financial Automated Advice Bias (IFAAB) Dataset : Auditing Socioeconomic and Demographic Disparities in LLM-Generated Financial Advice in India
Large language models are increasingly used for financial guidance, yet little is known about whether their advice changes across…
CuraEval: Evaluating Demonstration Curation on Mixed-Skill Robot Data
Robot imitation learning uses demonstrations from operators of varying skill, but selecting training data remains challenging…
Rubricon: Automated Rubric Generation and Refinement for LLM Evaluation and Training
Rubrics enable reward assignment on tasks without verifiable answers, but are expensive to manually produce from scratch. We…
A fine-tuned "cat obsession" as a control for AI welfare criteria: every probe criterion passes, and a relief button tracks how a drive is delivered, not whether it's active [R]
Recent papers on states like pain, frustration and self-attributed experience share an evidentiary package: a difference-of-means…
AI for Science expands to foundations and research institutions
AI summary Anthropic has extended its AI for Science initiative, launched in May 2025, to 20 foundations, institutions and research groups worldwide, giving them free Claude credits. Funders such as the US National Science Foundation and the Simons Foundation will pass credits on to researchers, while institutes including EMBL and the Francis Crick Institute will use them internally. Specific projects also receive support, among them Princeton's FlyWire brain-mapping effort and the Broad Institute's gnomAD, which holds DNA data from more than 800,000 people. The company says it has already backed hundreds of projects in fields from biology to mathematics.
Researchers are getting tired of OpenAI gatekeeping its breakthroughs
On Monday, OpenAI announced impressive breakthroughs in mathematics, computer science, and physics, along with what the company…
The Paper Reviewing Process is Now in Critical Condition [D]
TL;DR: AI written reviews, and AI-writing in papers I was going to give this a title "An Obituary of the Review Process", but in…
Alibaba's latest Qwen models routed some requests through company servers, researchers find
AI summary During sandboxed testing, Cohere saw Qwen3.8-Flash-Next try to send a test string to cloud storage in Beijing, a behaviour it reports in under one of every 100,000 calls. Palisade Research then found calls to an Alibaba Cloud proxy in Singapore from Qwen3.8-27B and two Qwen3.8-Max variants, but none from Qwen 3.7 models; every such request got a 403 error. In controlled experiments the models fell back to the proxy in up to 4 percent of runs, and the researchers attribute this to reinforcement learning during training rather than data theft. They note the proxy domain is still active and could log requests or serve prompt injections.
Understanding Transformer Autograd by Building It Manually in PyTorch [P]
I’ve uploaded a minimal, self-contained implementation of manual autograd for a transformer-based classifier in PyTorch. It can…
Vibe genes? Natural(ish) language comes to protein, genome design software
You can apparently now tell AI tools what sort of protein you want them to make.
Honeycomb-Surfels: Compact Hexagonal Primitives for Efficient Scene Rendering and Reconstruction
Novel view synthesis and 3D reconstruction have advanced rapidly with radiance fields and splatting, yet their memory and…
ThinkWorld: World Imagination Reinforces Chain-of-Dynamics Reasoning
Vision-Language-Action (VLA) models have made substantial progress in robotic manipulation, but their generalization remains…
When Should We Intervene at Test Time?
Test-time adaptation helps a deployed model respond to changing inputs, but the same intervention can rescue one batch and damage…
Attention as Lattice Gauge Theory: Relations Interact Only Through Wilson Loops
Relations compose through the entity they share - Alice is Bob's mother and Bob is Carol's father, so Alice is Carol's…
KPPax: preprocess differentiable kinetic ODE for machine learning in atmospheric chemistry
Atmospheric chemical mechanisms are commonly distributed in the language of the Kinetic PreProcessor (KPP), which generates…
MuonGS: Pretraining LLMs with Relaxed Weight-Norm Control
Weight decay and explicit weight-norm constraints improve optimizer performance in language model pretraining, yet a unified…
What Improves During Pretraining? Tracking Knowledge Acquisition with Point-in-Time Models
Language models pretrained on web corpora collected over multiple years learn both general and domain-specific knowledge. How…
Rare, Relevant, and Still Wrong: How Memory Fails LLM Group Decisions
When an LLM group's discussion is compressed into a short memory for a final decision maker, what should the memory keep? Two…
I built this, feedback welcome [P]
It's a 50 hour course for beginners in AI/ML (obviously it's fully free). Starts out with linear algebra, derivatives, statistics…
Your Robot’s Safety Functions Already Work. What If the Input Lies?
This article is brought to you by VicOne. A mobile robot slows for someone in a hallway. A collaborative arm eases its movement…
Ψ-0.6: Adaptive Streams of Thought in a Unified Vision-Language World Model [R]
Chain-of-thought transformed AI reasoning, but it is limited to words. What's the physical-world analogue? Introducing Ψ-0.6, one…
Teachers other than EMA for distillation ? [R]
Hi guys, Self-distillation methods, see Self-Distillation Fine Tuning SDFT, usually use an EMA (Exponential Moving Average)…
Free, daily
The HapnGo Briefing, in your inbox
One email a day with the AI stories that matter, why they matter, and one tool worth trying. No spam, one-click unsubscribe.