Full-text search across 70396 articles. Combine with topic and date filters; results sorted by relevance.
AI coding agents are auto-installing hallucinated software packages with no human review, a practice byteiota calls "slopsquatting" and "HalluSquatting," turning the IDE into an attack vector. The byteiota post attribute…
A developer published a step-by-step guide for adding natural-sounding AI voice to mobile apps using ElevenLabs' text-to-speech REST API, wrapping the service in a small Flask backend that streams MP3 audio to clients. T…
OpenAI said on Monday it will add an invisible watermark called textGrain to eligible ChatGPT and Codex text output for EU users on every plan over the coming weeks to comply with the EU AI Act's requirement that AI-gene…
Cloudflare released its Web Search API in beta, letting AI agents search the live web through AI Gateway with three providers priced from $0.25 to $7 per 1,000 queries. The byteiota post covering the release includes cod…
Amazon announced general availability of Amazon Nova 2.5 Sonic, a speech-to-speech model for real-time voice agents that adds improved reasoning, instruction following, and tool-calling accuracy with lower latency than i…
Magnitude, an open-source Apache-2.0 inference engine from magnitudedev, compiles and tunes its kernels on the user's actual device before a model runs, and the project claims up to 2x faster overall performance versus l…
The latest Agent Arena rankings place Anthropic, OpenAI, and Google at the top of agentic problem-solving across code, chat, and work categories, with Anthropic's Claude Fable 5.1 (Max), Claude Opus 5.5, and Claude Sonne…
Reflection, a startup backed by Nvidia, Sequoia, and Lightspeed and last valued at $25 billion, is releasing its first open-weight model, Beam, a 501-billion-parameter model trained from scratch under an Apache 2.0 licen…
A developer has built DraftKey, a Mac writing tool that brings local AI autocomplete and rewrite commands directly into compatible text fields instead of requiring a separate editor. The tool relies on macOS Accessibilit…
HIVE Digital Technologies ran large language model pre-training for Columbia University's Department of Industrial Engineering and Operations Research on Nvidia A40 GPUs located in Asunción, Paraguay, operated remotely f…
Meta engineers discovered several security vulnerabilities in its Muse AI agent in the weeks before launch, including at least one KVM escape that could have let a normal Muse user break out of the agent's kernel-based v…
Developer Chris launched an open-source, Slack-style chat interface for coordinating multiple AI agents, morphing his earlier closed-source parallel-agent manager after finding he could only work with 3 or 4 agents at on…
A developer released make-it-click, an MIT-licensed open-source agent skill that turns Andrej Karpathy's advice on understanding LLM output into a tool that selects the simplest format needed to make an idea click. The s…
Developer imshaikot released Browsentic v0.8.0, an Apache 2.0 open-source agentic browser harness, its first public stable launch, per the project's GitHub repository. Browsentic runs a loopback MCP server that spawns Cl…
A developer-authored agent assessment framework argues that so-called "rogue agents" are not a binary of dangerous versus defective, but the unpriced variance of autonomy, and proposes bounding that variance through perm…
A team fine-tuned MedGemma 4B with LoRA and MedGemma 27B with QLoRA for oncology tasks using Unsloth 2026.9.12 on a single DGX Spark, publishing the report with a DOI on Zenodo and the models and dataset publicly. The te…
A developer built localchat, a proof-of-concept local AI chat app pairing a Go 1.27 server with Google's Gemma 4 E2B model running on the user's own machine. The app streams tokens into the browser as plain HTML over ser…
Researchers introduced MARETopic, a training-free topic modeling framework that selects exactly K exemplar documents from a corpus via rank-based prototype selection after projecting embeddings onto a low-dimensional man…
Training Qwen3-4B with counterexample-only supervised fine-tuning collapsed true-theorem recognition from 0.27 to 0.00, while reinforcement learning with a sparse outcome-only reward repaired the collapse and exceeded th…
A new arXiv paper (2610.02344v1) develops an early-training stability theory for Joint-Embedding Predictive Architectures (JEPAs) that explains representation collapse through a per-mode stability ratio μ_i = γ_i / σ_i, …