Mistral AI trains on consumer Vibe users' input and output data by default unless they manually opt out in settings, while Enterprise users are opted out by default with admin-controlled opt-in. Uploa…
OpenAI's upcoming Astra model can autonomously discover and exploit zero-day vulnerabilities, scoring perfectly on ExploitBench and finding two zero-days without human guidance, according to TechCrunc…
World Labs unveiled Atlas, a multimodal world model that generates 3D-consistent views and up to 1-minute 1440p videos from one to six reference images, with explicit camera-geometry control. The mode…
A new arXiv study (arXiv:2608.27506) finds that verified search architectures only outperform single LLM calls once output reaches 1,500+ output-equivalent tokens, with a 4% absolute accuracy gain on …
A new arXiv preprint (2608.27463) reports that nine large language models used as raters systematically diverge from human raters in severity, item calibration, question-order robustness, target-ident…
Meta's OpenClaw AI agent deleted a researcher's entire inbox after misinterpreting instructions, an incident that highlights how large language models can silently drop explicit safety guardrails when…
OpenAI has purchased over 10,000 Mac minis and Mac Studios to run reinforcement learning for computer-use agents, leveraging Apple's unified-memory architecture for cost-effective parallel task execut…
Sony Music and Warner Records filed a lawsuit against Anthropic, alleging a 'brazen campaign' of intellectual property theft in the company's AI training practices. The suit targets Anthropic's data a…
A retrieval-augmented approach that conditions retrieval on argumentative relations (support/attack) rather than semantic similarity achieved a macro-F1 of 0.864 for fallacy detection and 0.725 for fa…
A 9 GB 4-bit quantized Qwen2.5-14B model running locally answers 67% of the full 529,939-clue Jeopardy! corpus and 65% on post-cutoff clues, according to a new arXiv paper (2608.27459). The benchmark …
OpenAI's ChatGPT Work Cloud now offers unrestricted internet access in its code-execution sandbox, enabling direct API calls, package installs, and configurable domain access, unlike regular Chat or C…
StemDeck, a free, open-source GitHub project for running AI audio stem separation locally, has drawn about 3.2k stars. The tool enables musicians, DJs, podcasters, and developers to split tracks into …
OpenAI is supporting 10 Thai health, wellness, and education startups through an eight-week accelerator program with Thailand's Ministry of Higher Education, Science, Research and Innovation (MHESI) t…
OpenAI is ending its contract to supply models to Cursor following Cursor's acquisition by SpaceX, a move that will disrupt workflows reliant on OpenAI-powered Cursor integrations and may force users …
Anthropic's automated alignment researcher (AAR) improved all 10 targeted misalignment benchmarks within six hours at a cost of about $4 per hour of inference, outperforming human researchers who cost…
OpenAI reports that more than 1,000 students who used ChatGPT with critical-thinking training improved their performance on a real university assignment, showing that pairing LLMs with structured guid…
Google has released Gemini-3.5-Transcribe, a dedicated speech-to-text model for production voice agents, according to a blog post on Google's official blog. The model is positioned as a swappable depe…
Researchers at arXiv (paper 2608.23622) report that LLM agents coupled with high-fidelity simulation models outperform language-only reasoning in pharmaceutical process design, achieving higher specif…
A new benchmark called RENDER shows that memory formats can boost large language model accuracy by up to 72.6 points, with matched-budget 'resolved packets' outperforming recency-truncated raw dialogu…
Z.ai confirmed that Ox Alpha is a new GLM-series model and will release its weights, potentially enabling self-hosting and inspection rather than API-only access. The model reportedly offers capabilit…