Research news and analysis on Web Pulse: 583 curated articles tracking the latest Research developments, tools, and research, updated continuously from vetted sources.
In Reinforcement Learning with Human Feedback (RLHF), human preference collection involves generating multiple responses for a given prompt and asking people to select the preferred one, which is faster than manual respo…
A benchmark called "1rok" that evaluates seven frontier large language models (LLMs) as stock pickers, giving each $100,000 in paper capital and identical tools to select stocks weekly. The project, which started on Janu…
Rich Thau, a veteran public opinion researcher and president of the consulting firm Engagious, has joined the Foundation for Individual Rights and Expression (FIRE) Advisory Council. Thau is known for creating and modera…
The first draft of the "Procedural Fascicle," part two of the R7RS-Large Foundations, has been released by Working Group 2, covering familiar block programming forms like `lambda`, `let`, `if`, `or`, and `set!`. Its majo…
The author successfully connected an AI agent called OpenClaw to a physical robot arm, enabling it to configure the arm, grasp objects, and train another AI model for object manipulation. This experiment highlights the "…
ClawBattle, a new open-source benchmark that tests multimodal LLMs on code golfing tasks, which require both visual and textual understanding. It claims the benchmark avoids data contamination by using confidential, top-…
Google Beam has introduced a new experiment that uses HP Dimension's immersive display to render remote participants in true-to-life size and position them as if seated around a table, paired with spatial audio to anchor…
Mathematical and network logic behind achieving perfect first-shot accuracy in Valorant after a clean movement key transition. It details two distinct timing windows—a client-side grace period of 15-38 milliseconds and a…
The Gemma 4 26B model was the first local AI to achieve a perfect score on the HumanEval benchmark, including solving the notoriously difficult problem 145. This problem requires sorting integers by the sum of their digi…
Nobel laureate Olga Tokarczuk initially appeared to admit using AI to assist in writing her latest novel, stating she asked a language model for song titles and creative development suggestions. However, she later denied…
Stable Audio 3 is a family of fast latent diffusion models (small, medium, large) designed for variable-length audio generation and editing, capable of producing several minutes of sound. The models use a novel semantic-…
The article describes Wi-Wi STAMP, a wireless time synchronization protocol developed by Japan's NICT that uses 900 MHz band interferometry to achieve picosecond-level phase sync and nanosecond-level time sync, with prot…
Stability AI has released Stability Audio 3.0, a new family of audio models that includes a large model capable of generating professional-grade music compositions over six minutes long. The company is making the small a…
DEPOSE, a cryptographic forensic framework designed to create verifiable audit trails for AI coding agent sessions (like Claude Code or Codex CLI). It addresses the problem that standard session logs (JSONL files) are no…
Methodology for testing distributed and stateful systems using AI coding agents that execute two specialized skills: one to design a claim-driven test plan and another to execute it. The approach produces structured Mark…
In April 2026, DeepSeek released its V4 model, a 1.6 trillion parameter MoE architecture, and for the first time officially validated its inference on Huawei's Ascend 950PR chip, marking a significant milestone for China…
"A Fact A Day," an autonomous daily educational podcast that summarizes scientific and technological breakthroughs in about 60 seconds. The entire production process, from researching sources like MIT Tech Review to scri…
At Google I/O 2026, Google announced that its AI Mode surpassed one billion monthly users within a year of launch, with query volume doubling every quarter. The milestone was accompanied by two major structural changes: …
Interactive visual guide to LeNet-5, a neural network designed to recognize handwritten digits. It explains key components like convolution and pooling layers, feature maps, filters, and the RBF output layer, allowing us…
The author ran three Claude Code sub-agents (an explore-reviewer, security-reviewer, and plan-architect) on the same 500-line pull request and found that the agents disagreed on 41% of their 78 total comments, with each …