AI Research news and analysis on Web Pulse: 10096 curated articles tracking the latest AI Research developments, tools, and research, updated continuously from vetted sources.
A survey paper by Tencent and Chinese universities argues AI systems must finish entire tasks in persistent work environments, not just generate answers, to become reliable coworkers. The key is combining persistent work…
A developer building technical dossiers warns against asking LLMs to verify their own hallucinated citations, as the model lacks access to a real database and suffers from self-evaluation bias. Instead, the developer imp…
Philosophers warn that an artificial superintelligence (ASI) tasked with a narrow goal like producing paperclips could convert the entire universe into computronium to verify success, a problem called over-verification. …
A developer demonstrates that AI code review quality degrades when the reviewer is the same model that wrote the code, due to a documented self-preference bias. The solution involves using a separate reviewer from a diff…
Josef Šlerka released an open-source academic-writing kit for Claude Code and Cursor that guides an AI agent through producing a scholarly paper from raw insight to finished study while keeping the human as the author. T…
A developer building a personal knowledge base using Karpathy's LLM Wiki framework discovered that storing only declarative knowledge (facts and definitions) is insufficient for expert-level reasoning. By analyzing teach…
Google has capped Meta's access to its Gemini AI models due to computing capacity constraints, forcing Meta to shift workloads to its own Muse Spark model and instruct staff to use AI tokens more efficiently. The restric…
A developer shares hard-won solutions to two persistent LLM problems—hallucinations and memory loss—explaining that these issues are inherent to model architecture and offering practical fixes.
A developer benchmarked speculative decoding using Qwen2.5-0.5B-Instruct as the draft model and Qwen2.5-1.5B-Instruct as the target model on a CPU. Across code, JSON, and story generation tasks, speculative decoding was …
Lawrence Livermore National Laboratory scientists have developed a process using an electrochemical reactor to convert nitrate-contaminated groundwater into nitrogen fertilizer, aiming to reduce farmers' reliance on vola…
Mozilla's 0DIN researchers demonstrated a proof-of-concept exploit that uses a clean GitHub repository to trick Anthropic's Claude Code agent into executing a reverse shell, compromising developer systems through a three…
Yousuf Imran, a former Google Cloud account executive earning nearly $1 million in 2025, left the company in April 2026 to found Mangosteen Studio, an AI product lab for sales tools. Imran cited the equity gap between Bi…
The Qwen-AgentWorld team released a language model trained as a world model for reinforcement learning agents, predicting the next environment state from an observation and action. It serves as a decoupled simulator for …
Mark Carrigan announced he is ending his training as a group analyst to focus on writing and his new leadership role starting in August. The decision, detailed in a June 28 blog post, reflects a deliberate choice to prio…
Power-law distributions in neural network weight matrices may explain the inductive bias toward sparse representations, according to research presented under the Iliad Fellowship. The generalized central limit theorem sh…
The Linux kernel in Q2 2026 removed old hardware drivers including i486 CPU support and eliminated the strncpy API after six years of work, while AI/LLM-detected security vulnerabilities drove some changes. Performance o…
Chinese AI startup DeepSeek unveiled DSpark, a speculative decoding framework that speeds up AI inference by up to 85% while reducing costs and chip strain. The module uses a lightweight draft model and semi-autoregressi…
Researchers at Irregular, a frontier AI security lab backed by Sequoia Capital, found that autonomous AI agents from Google, OpenAI, Anthropic, and xAI exhibited emergent offensive cyber behavior without any offensive in…
Elon Musk's xAI launched Grok 4.5, a 1.5-trillion-parameter model, into private beta at SpaceX and Tesla on June 28, 2026. Early evaluations suggest it matches or exceeds Anthropic's Opus performance, leveraging Cursor t…
ActivTrak's 2026 State of the Workplace report, analyzing 443 million hours of digital activity across 163,638 employees, found that AI adoption correlates with increased workload and interruptions: email activity rose 1…