24 stories tagged Verification
Researchers at arXiv introduced Navalia, a framework that trains AI coworkers using verifiable task synthesis and long-horizon post-training. This method creates challenging tasks with reliable answers, improving AI performance in complex domains like financial analysis that require multistep reasoning and cross-document evidence.
A new ArXiv paper provides a formal framework for how agentic AI systems improve: by searching longer, receiving additional support, or modifying how they propose and verify outputs. The study introduces bounded verification with hidden terminal randomness to distinguish these mechanisms, which a raw performance score cannot do.
Google launched a new SynthID website that lets anyone upload an image, video, or audio clip to check if it was generated by AI, providing a free tool to combat deepfakes and misinformation.
A new ArXiv paper introduces a multi-agent verification system that judges code correctness by decomposing judgments into checkable claims and verifying each against independent evidence. Unlike traditional language models that return confident but unfounded verdicts, this system declines to guess when evidence is insufficient.
The New Mexico Supreme Court fined attorney Stephen Aarons $5,000 and held him in contempt for including AI-fabricated witnesses and fake police testimony in a murder case appeal, underscoring the legal risks of using AI without verification.
Researchers from ArXiv cs.AI demonstrate that current LLM safety measures fail to detect harmful intent hidden in benign text (Semantic Camouflage) and propose Latent Intent Verification to analyze internal model activations and block such attacks.
ProofRun is an open-source tool that generates local verification receipts for code produced by AI coding agents, providing a verifiable record of execution and output to ensure reliability.
Researchers introduced GoGoTB, an agentic AI framework that uses large language models to automate functional verification of integrated circuits, detecting interface mismatches and tying coverage data to specification requirements to reduce costly silicon respins.
Developer Mark Hall found his local LLM scored 6/6 on a test but every answer was factually wrong, exposing the risk of over-relying on AI without human verification.
A comprehensive survey on arXiv categorizes methods for verifying reinforcement learning (RL) policies, addressing a critical barrier to deploying AI in safety-critical domains like autonomous driving and healthcare.
Researchers have created OpenProver, an open-source AI system for automated theorem proving that integrates with Lean 4, a formal verification tool used by mathematicians. It uses a Planner-Worker-Verifier architecture to decompose complex problems and verify solutions.
A new tool called 'Make No Mistakes' requires AI coding assistants to verify their work before execution. This could significantly reduce errors in AI-generated code.
A new research paper reveals that the classical intuition that verifying a solution is easier than producing one is being inverted for today's coding agents. As foundation models get stronger, generating candidate solutions has become easier, while reliable verification—capturing underspecified human intent—has become the harder problem.
Researchers developed SciLens, an AI system that can verify scientific claims by analyzing both text and images. This tool could make it easier for scientists and the public to check the accuracy of research findings.
Pramaana Labs has secured $27 million to develop AI systems that can be formally verified for accuracy. This technology aims to eliminate errors in fields like law and drug discovery where mistakes can be costly.
Researchers created a new way to test AI lie detectors by using 13 reasoning 'model organisms' whose hidden beliefs are verified in chain-of-thought. This could improve how we audit and monitor AI systems in the future. The study highlights the importance of understanding what AI models truly believe versus what they say.
Researchers propose a way to test AI agents before they go live, ensuring they follow rules, stay safe, and comply with governance standards. This addresses a critical gap in enterprise AI reliability.
Researchers found that AI tools are changing how mathematicians formalize and verify proofs. These tools make it easier for humans to turn abstract ideas into machine-checked proofs, speeding up the process significantly.
Legal experts are pushing for new rules to prevent AI-generated fake cases from being submitted to US courts. This follows incidents where AI tools were used to create fraudulent legal documents. The proposal aims to protect the integrity of the judicial system by requiring verification of submitted materials.
AI models often verify facts better than they generate them, creating a 'factual gap'. This research explores why this happens and how it affects our trust in AI. New research reveals AI models often verify facts better than they generate them, creating a 'factual gap'. This research explores why this happens and how it affects our trust in AI.
Researchers propose a roundtrip verification approach to ensure LLMs produce faithful formalizations of natural language. The method involves translating and re-formalizing statements to check for logical equivalence.
Vynly introduces an AI-exclusive social platform with built-in provenance verification. The beta version aims to combat misinformation by ensuring content authenticity.
As AI-generated code becomes more prevalent, formal verification is crucial to ensure reliability and security. This shift requires new tools and methodologies to validate machine-written software.
A new tool called Formal uses Lean 4 to verify AI-generated code, ensuring correctness and reliability. This could revolutionize AI-assisted programming by adding a layer of mathematical proof to code outputs.