The Archive

Question about score variance across reviewer batches at ICML 2026 and peer review calibration.

LIDARLearn: open-source PyTorch library for 3D point cloud learning with 56 model configs, cross-validation, and PDF publication generation.

Reflection on OpenClaw developments with minimal substantive detail provided.

Blog-to-newsletter tool enhanced with new content type via agentic prompt engineering patterns.

BabyZWM demonstrates zero-shot visual-cognitive competence from single-child dataset, narrowing data efficiency gap between AI and human development.

PyCon US 2026 announcement with new AI and security tracks, May 13-19 in Long Beach.

Coding agents are starting to write production code at scale. Stripe’s agents generate 1,300+ PRs per week. Ramp attributes 30% of merged PRs to agents.... Coding agents are starting to write production code at scale. Stripe’s agents generate 1,300+ PRs per week. Ramp attributes 30% of merged PRs to agents. Spotify reports 650+ agent-generated PRs per month. Tools like Claude Code and Codex make hundreds of API calls per coding session, each carrying the full conversation history. Behind every one of these workflows is an inference stack under… Source

World, which has raised eyebrows (but also a lot of interest) with its Orb-centered anonymous verification project, is looking to expand its influence via a bevy of new partnerships.

Last month, OpenAI gave up on its Sora video generation tool, and on Friday, the Sora team's leader, Bill Peebles, announced that he is leaving the company. OpenAI has been shifting its priorities as part of an effort to avoid "side quests," and Peebles' departure is just one of many recent changes as the company moves to focus more on coding and enterprise use. As part of a note Peebles posted on X, he said: I am immensely grateful to Sam, Mark, Aditya and Jakub for fostering a research environment that allowed us to pursue ideas off-the-beaten path from the company's mainline roadmap. It's ...

Tinder users who prove they're a real person by visiting an identity-verifying orb will soon be able to get five free boosts in the app - and it's just the latest service to embrace the orb. World, which was co-founded by OpenAI CEO Sam Altman, initially tested Tinder verification using its facial scanning orbs through a pilot program in Japan last year. It's now expanding the service to "select markets, including Japan and the United States." To verify that they're not a bot or an AI agent, users have to physically visit one of World's orbs in person. According to World, the orb "takes pictu...

Kevin Weil and Bill Peebles are leaving OpenAI as the company shuts down Sora and folds its science team, signaling a sharp pivot away from consumer moonshots toward enterprise AI.

The Trump administration has spent nearly two months fighting with AI company Anthropic. It's dubbed the company a "RADICAL LEFT, WOKE COMPANY" full of "Leftwing nut jobs" and a menace to national security. But some of the ice may reportedly be melting between the two, thanks to Anthropic's buzzy new cybersecurity-focused model: Claude Mythos Preview. Anthropic's relationship with the Pentagon soured quickly in late February after the company refused to budge on two red lines: using its technology for domestic mass surveillance or lethal fully autonomous weapons with no human in the loop. Ant...

Returning backers a16z and Thrive are expected to lead the round.

Agents are evolving from question-and-answer systems into long-running autonomous assistants that read files, call APIs, and drive multi-step workflows.... Source

There's a lot more code — but it's a lot more expensive and requires a lot more rewriting.

Data centers face construction delays and energy bottleneck as resistance grows.

ASMR-Bench evaluates AI auditor detection of subtle sabotage in ML research codebases via 9 codebases with intentional flaws in hyperparameters, data, or evaluation logic.

Method for learning low-dimensional manifold structure in stochastic dynamical systems via geometric regularization of autoencoders using ambient covariance.

Knowledge Graph-LLM system for manufacturing ML explainability: domain data + ML results stored in structured KG with selective LLM-based retrieval.

Benchmark suite for LLM-based drug design via chemically-grounded RL environments: molecular property prediction, representation, and design tasks.

DeepInsightTheorem framework trains LLMs for informal theorem proving by teaching recognition of core problem-solving techniques via hierarchical reasoning.

Cross-linguistic study of politeness effects on 5 LLMs (Gemini-Pro, GPT-4o Mini, Claude 3 Sonnet, DeepSeek-Chat, Llama 3) via 22,500 English/Hindi/Spanish prompts.

I kind of wish it just took pictures. I've never been as charmed and frustrated by one gadget as I have with the Poetry Camera. It's a delightful object. White and cherry red with a color-matched woven strap, it looks playful and adorably lo-fi. If I saw it on a store shelf, I'd absolutely pick it up. But aside from obviously appealing, I'm not exactly sure what it is. I mean, I know what it is. It's a camera that makes AI poems instead of photos. You take a picture, and instead of printing a photo, you get an AI-generated poem inspired by the scene, printed on thermal receipt paper. But afte...

VEFX-Bench: large-scale video editing benchmark with human annotations, complete examples, quality labels, and specialized evaluation metrics for instruction-guided editing.

Dual-aspect evaluation framework for 4 LLMs (GPT-4o, Claude 3 Opus, Gemini 1.5 Pro, Grok-1) on Vietnamese legal text simplification: accuracy, readability, consistency.

Deep learning workflow for joint flood-landslide susceptibility mapping via spatial partitioning, probabilistic Early Fusion, Late Fusion, and Mixture of Experts ensemble.

Information Router architecture mitigates modality dominance in VLMs by routing vision-language inputs based on signal-to-noise ratio rather than attention alone.

SwanNLP proposes LLM framework with structured reasoning for plausibility scoring in narrative word sense disambiguation at SemEval-2026 Task 5.

Study empirically compares task-reward RL against distribution sharpening, showing frontier models develop new agent capabilities beyond latent skill elicitation.

Cross-domain empirical study characterizes LLM-generated competency questions for ontology engineering across open and closed models.