AI Labs Pulse

Product Launches

• 🔥 xAI Voice Agent API: xAI launched the Voice Agent API for real-time voice conversations over WebSocket, billed at a flat...

• Meta's $2B acquisition of Chinese-rooted AI startup Manus was billed as a victory for Zuckerberg.
• China responds with punitive actions against...

New tools spotlight persistent AI limitations amid AGI hype:

• Cognitive framework rethinks AGI measurement (71 HN pts).
• Oboe pushes custom LLM...

Qianfan-OCR emerges as a unified end-to-end model for document intelligence, advancing enterprise OCR capabilities – paper and apps now available for exploration.

Major LLMs exhibit demographic biases in hiring evaluations:

• Base models (e.g., Copilot) advance elite affiliations/connections—even skipping 3-year...

New paper 'Thinking in Uncertainty' introduces Latent Entropy-Aware Decoding for mitigating hallucinations in MLRMs. Join the discussion on this breakthrough.

Pricing evolution underway as OpenAI deems current unlimited model accidental and unsustainable amid 900M weekly users and 50M subscribers.

-...

Handy dev tool for Claude Code power users:

• Live macOS menu bar display of 2x usage window status, duration, and switches.
• Ends the pain of manual...

Microsoft has hired the full team from Sequoia-backed AI collaboration startup Cove, which is shutting down with service ending April 1 and customer data deletion planned. A strategic talent grab for AI tooling.

MiroThinker-1.7 & H1 pushes boundaries with verification techniques enabling heavy-duty research agents. New paper details this breakthrough in agentic capabilities.

US DOD labels Anthropic an 'unacceptable risk to national security', citing fears it might "attempt to disable its technology" during warfighting...

OpenAI pushes agentic AI with two new models: GPT-5.4 Mini built for agentic workloads.

Joint paper from OpenAI, Anthropic, DeepMind, Meta shows AI...

One-Eval launches as an agentic system for automated and traceable LLM evaluation, streamlining benchmarking with built-in traceability. Join the discussion on this paper.

Key strategies from Amazon AI leader Abhinav Kasliwal to scale GenAI prototypes 3x faster with safety:

• 5 Pillars: Design, Evaluation, Safety,...

Emerging trend: LLM agents automate post-training and runtime learning, enabling faster iteration without weight updates.

• PostTrainBench tests...

• US Treasury's pragmatic playbook introduces 230-point AI Risk Management Framework for financial institutions.
• FinToolBench evaluates LLM agents...

Ultra-simple tool launches autonomous AI agents with sandboxed execution in just 2 lines of code – buzzing at 48 points on Hacker News, perfect for rapid prototyping.

Mistral AI releases Forge, drawing huge buzz with 598 points on Hacker News – a hot new product launch to watch.

A new methodological review systematically identifies and categorizes control conditions, including placebos, used in interventional studies of LLM-based chatbots – essential for boosting rigor in AI experiments.

Gemini 3.1 Pro is reshaping AI evaluation:

• Benchmark obsolescence: Experts question if traditional scores measure real intelligence.
• Vibe Era...