AI & Machine Learning

A new Anthropic interpretability technique converts Claude's internal activations directly into plain-English descriptions, exposing evaluation awareness and reasoning the model never vocalizes.

Qwen3.7-Max debuts at Alibaba's Cloud Summit in Hangzhou as the company's most capable agent model, sustaining continuous autonomous execution for 35 hours in internal testing.

Andreessen Horowitz leads a $250M Series C in Exa, valuing the AI-native search startup at $2.2B as it serves over 5,000 companies including Cursor, Cognition, and HubSpot.

Google's new efficiency flagship outperforms Gemini 3.1 Pro on most evals while running 4x faster and costing 40% less.

At Google I/O 2026, Google previewed Android XR smart glasses powered by Gemini 2.5 Pro with partners Samsung, Warby Parker, Gentle Monster, and XREAL, while reaffirming the summer rollout of Gemini Intelligence across its device fleet.

Inceptio and Pony.ai executives push back on hype: LLM advances don't translate to AV commercialization, which still depends on accumulated miles, regulation, and industry partnerships.

The four-year commitment — described as the largest deal of its kind between an AI company and a global philanthropy — targets health services for 4.6 billion people in low-income countries.

A Sinch survey of 2,527 AI decision-makers finds 74% have shut down live AI customer communications agents, with rollback rates rising to 81% among the most governance-mature organizations.

Cloudflare detailed the custom infrastructure behind its Workers AI expansion to host models like Kimi K2.5, including a Rust-based inference engine and a lossless weight-compression system.

Nine days after v0.13 closed Hermes Agent's durability gap, Nous Research shipped v0.14 on May 16 — adding pip install from PyPI, native Windows, and a local OpenAI-compatible proxy that wraps Claude Pro, ChatGPT Pro, and SuperGrok subscriptions.

After an Alpha Omega-mediated Mythos scan turned up five claimed vulnerabilities, the curl security team triaged the list to one low-severity CVE bound for curl 8.21.0 in late June.

A University of Pennsylvania team led by Vivek Shenoy has published Mollifier Layers in TMLR, a module that swaps recursive automatic differentiation for convolutional smoothing to make physics-informed neural networks stable on high-order, noisy inverse PDEs.