certification

This paper proposes a post-hoc certification framework for sparse autoencoder (SAE) based interpretability, deriving an upper bound on the frozen language model's risk using measurable quantities. The framework is validated on GPT-2 Small, Gemma-2B, and Llama-3-8B, showing non-vacuous bounds and revealing depth-dependent behavior.

CARVE-Q introduces a quantum-AI search layer for certified interactive driving repair, using quantum minimum finding on repair lattices while keeping safety authority classical. It provides structured certificates for vetoed maneuvers, achieving 100% right-of-way respect and blame consistency on INTERACTION replay scenarios.

Xiaomi announces that its flagship display, featuring XiaomiVisionCare, has achieved the world's first TÜV Rheinland Quadruple Eye Care certification, emphasizing its commitment to eye care technology.

CARVE is a certification framework for autonomous driving that provides runtime proofs for multi-agent repairs, accepting 98.64% of initially vetoed maneuvers on INTERACTION replay episodes without requiring predictions of other drivers' compliance.

This paper studies the exact certification problem for neural networks, showing that even minimal overparametrization can make certification exponentially hard for threshold circuits of depth≥2 and log-precision Transformers. It also characterizes approximate certification, revealing that allowing polynomially many mistakes still requires exponentially large certificates.

This paper proposes claim-selective certification for high-risk medical retrieval-augmented generation (RAG), decomposing responses into verifiable claims and scoring them against evidence to produce actions (full, partial, conflict, abstain) using an intent-aware selector, achieving low unsupported-claim risk and high action accuracy.

A full-day hands-on bootcamp on May 30 for Claude Code, featuring 10 real projects, certification from Packt Publishing, and curriculum covering CLAUDE.md setup, prompt patterns, and Git workflows.

GitHub has announced a new official certification for 'agentic AI developer' (exam GH-600), testing skills in multi-agent orchestration, state management, and system design. The certification, backed by GitHub and Microsoft, will be generally available in July 2026, with early access at 80% off for the first 100 beta takers.

GitHub has introduced an official certification called GH-600 for the Agentic AI Developer role, recognizing the discipline of operating and integrating AI agents across the software development lifecycle. This credential is expected to become highly valuable as demand for such experts grows.

Introduces the State-Centric Decision Process (SDP), a runtime framework that enables language agents to construct task-induced state spaces, observation-to-state mappings, certified transitions, and termination criteria from raw text environments, achieving state-of-the-art training-free results on five benchmarks.

OpenAI launches its first certification courses, including AI Foundations and ChatGPT Foundations for Teachers, with a goal to certify 10 million Americans by 2030. The AI Foundations course is available directly within ChatGPT through pilot programs with major employers like Walmart, Lowe's, and Accenture.

freeCodeCamp is an open-source learning platform offering free certifications in web development, machine learning, and languages, with interactive coding challenges and projects.

Anthropic announces the Services Track and Partner Hub within the Claude Partner Network, a tiered program to help enterprises identify qualified partners for deploying Claude in production, with over 40,000 firms having applied and major consultancies like Accenture, Deloitte, and PwC building practices around Claude.