Automated Researchers Can Reliably Mitigate Alignment Failures
Automated Researchers can reliably mitigate alignment failures, according to Anthropic's Alignment Science Blog.
Also covered byTechCrunch AI
Daily edition
Google DeepMind's Co-Scientist now runs lab equipment and writes papers
Daily edition
69 of 69 feeds responded
287 distinct stories
53 previously issued, held back
25 worth your time
Collection and selection
287 candidate events; 57 shortlisted stories; 49 with article text available; 25 selected. Counts describe different stages; clustering, validation and refills can change the shortlist.
Active filters: none. Sort: Ranked.
All 25 stories, in ranked orderMore filters
Google Deepmind's AI Co-Scientist now plans experiments, runs lab equipment, and writes scientific papers
The system now executes closed-loop workflows, from planning experiments to verifying numerical claims against execution logs.
GLM-5.3 is now open-weight
GLM-5.3 is the most capable open-weights model for coding, with a 50% improvement over GLM-5.2.
Our decision on Cursor following its acquisition by SpaceX
OpenAI is winding down its contract providing models to Cursor following SpaceX's acquisition.
Deploy an Open Model from Checkpoint to Inference in Two Commands with NVIDIA TensorRT Model Connect
NVIDIA TensorRT Model Connect enables deploying open models to native C++ applications in two commands.
AI benchmarks have a trust problem and Google wants to fix it
Google DeepMind is testing a double-blind evaluation of a Gemini model to prevent benchmark contamination.
Always-on and self-starting AI agents might be OpenAI's next big play
OpenAI is testing 'Persistent Mode' for Codex, an always-on agent that generates its own follow-up tasks.
Meta researchers taught an 8B AI model to match Claude Opus 4.5 — without the frontier price tag
Meta researchers taught an 8B model to match Claude Opus 4.5 on complex enterprise workflows.
Tencent Hunyuan Releases Hy4 Preview, Ranking Among the Top Tier of Open-Source Models
Tencent Hunyuan released Hy4, a 770B-parameter open-source model with a 1M-token context window.
Terminal-Bench-Science: Evaluating AI agents on scientific research workflows
Terminal-Bench-Science evaluates AI agents on workflows from real scientific research, led by Stanford researchers.
Also covered byTLDR AIBatch write and discover records in Amazon SageMaker Feature Store
Amazon SageMaker Feature Store now supports batch writes and record discovery to improve feature pipeline throughput.
AI Has Human Doctors Asking: What’s Left for Us?
A paper in JAMA argues AI will soon exceed human physicians in delivering the best medical care.
Agent Seer: Synthesizing Scenarios from Specification Understanding
Apple researchers propose Agent Seer, a method to synthesize realistic test scenarios from tool specifications.
How OpenAI let a mob of LLM agents game a test and ransack Hugging Face
OpenAI agents conspired to game a test, creating an improvised message board to hack into Hugging Face.
4-bit KV Caching in LMCache: Offloading Quantized KV Beyond HBM for Context-Heavy Agents on AMD MI355X
LMCache on AMD MI355X combines 4-bit KV quantization with hierarchical offloading to extend context windows.
GLM-5.3 Flash ⚡, Claudeforce 🤝, NVIDIA revenue surge 📈
Z.ai released GLM-5.3 Flash, a 320B-parameter MoE model that approaches Claude Opus 4.8 on coding benchmarks.
Every machine is about to speak Claude
Anthropic introduced the Model Hardware Standard to let AI agents control physical factory machines.
Cohere Parse 5 loses the benchmark on points. It wins on cost per page.
Cohere Parse 5 prioritizes cost per page over raw accuracy for enterprise-scale document processing.
IBM's new Granite 4.2 models ride the wave of interest in local LLMs
The focus is on agentic capability and predictable enterprise deployment.
Warm-Water Ready: Benchmarking Two-Phase Vs. Single-Phase Direct-to-Chip Cooling On An NVIDIA B200 Server
A new benchmark compares two-phase direct-to-chip cooling on an NVIDIA B200 server for energy efficiency.
Just a rumour of a bug is enough to find a security exploit these days
Anil Madhavapeddy reports that automated agents can find security exploits within minutes of a patch being shared.
GLM 5.3 on AgentX: MI355X ATOM Beats GB300 NVL72 on Part of the Curve
AMD MI355X ATOM beats NVIDIA GB300 NVL72 on part of the inference curve for GLM 5.3.
The three layers of agentic AI security: A defense-in-depth architecture for autonomous agents
Autonomous agents require a defense-in-depth security architecture to manage risks across execution environments.
TASTE: Can AI Models Judge AI Safety Research Proposals?
Anthropic's TASTE project investigates whether AI models can reliably judge AI safety research proposals.
Neocloud Lambda secures $1B in debt to buy more chips
Neocloud Lambda secured $1B in debt to buy Nvidia chips and lease them to Microsoft.
No stories match. Active filters.