AI Research news and analysis on Web Pulse: 9956 curated articles tracking the latest AI Research developments, tools, and research, updated continuously from vetted sources.
Zenith, a new agent harness developed by an unnamed team, lifted OpenAI's GPT-5.5 from 5th to 1st place on the FrontierSWE benchmark, outperforming Anthropic's Claude Fable. The harness improves performance by optimizing…
DeepReinforce released Ornith-1.0, an open-source family of coding models under MIT license on June 25, designed specifically for AI agents operating in real terminal and repository environments. The 9B variant achieves …
A replication of the 2026 paper 'Prompt Injection as Role Confusion' on a single consumer GPU confirms that style-based prompt injection attacks work, but causal tests using activation steering and patching fail to deter…
OutYet.ai, a platform tracking upcoming AI model releases, lists eight models from major labs including OpenAI, Anthropic, and Google DeepMind, with most still unconfirmed or in development. Among them, GLM-5.2 is public…
Physicists have noted that Avogadro's number and the number of stars in the observable universe are roughly equal, prompting questions about whether this is a coincidence or has a deeper anthropic explanation. A new anal…
A developer used Comma.ai's open dataset to build a case study on Automatic Emergency Braking (AEB) systems, demonstrating how the data can be leveraged for safety-critical automotive applications. The project highlights…
Researchers introduced Simplified Sparse Attention (SSA), a method that uses gist tokens to achieve sparse attention without architectural changes, outperforming baselines on LongBench and improving retrieval-augmented g…
A quasi-mathematician offering services in mathematical concept translation and LLM behavior optimization has posted on Hacker News, inviting collaboration via email and GitHub.
AI democratization, while beneficial, creates a disconnect between deep understanding and widespread use, mirroring the pattern seen in past technologies like electricity and speech recognition. The author argues that as…
Qihoo 360 founder Zhou Hongyi unveiled Tulong Feng, an AI vulnerability agent, at ISC.AI 2026, claiming it as China's answer to Anthropic's Mythos. Separately, Z.ai released GLM-5.2 under an MIT license, which outperform…
Nvidia will present breakthroughs in neural rendering, world models, and simulation at its SIGGRAPH 2026 keynote on July 20 in Los Angeles. The session, led by VPs Jan Kautz and Ming-Yu Liu, covers AI-driven advances for…
A new analysis suggests that under illusionist theories of consciousness, AI systems may become conscious above a certain complexity threshold but could surpass the need for consciousness once sufficiently advanced. The …
Reloadium launched an AI-powered investigation tool that transforms any query into a structured matrix of entities, events, and relationships using Gemini AI and Google Search grounding. The platform provides confidence …
Executives report that AI is not yet delivering ROI but remain committed to figuring it out, according to Garrett Lord. The article argues that evaluations (evals) will become the strategic intellectual property defining…
Researchers introduced WGO-Bench, a benchmark for testing robotics subtask annotation performance across 100 egocentric and robot-video episodes with 743 annotated segments. After over 60 experiments, they found Gemini m…
Enterprise infrastructure faces bottlenecks when scaling large language models, as accelerators often idle due to data starvation from traditional storage protocols. The SNIA presentation at AI Infrastructure Field Day 5…
Mixedbread Search introduced asymmetric quantization for late interaction retrieval, storing document vectors as 1-bit binary signs while keeping query vectors at higher precision, achieving a 32x reduction in storage pe…
Meta's FAIR lab, in collaboration with Spain's BCBL, unveiled Brain2Qwerty, a non-invasive AI system that decodes typed sentences from brain activity with up to 80% character accuracy using magnetoencephalography (MEG). …
Atlarix, an agent workstation for open-weight models, resolved 42 out of 89 tasks on Terminal-Bench 2.0 compared to opencode's 39 out of 89, using the same model and identical infrastructure. The 3-task difference falls …
PriEco, a new open source web search engine, has been announced. The project emphasizes quality, faithfulness, consistency, and robustness, built over time rather than overnight. The creator advocates for physical note-t…