Diogo Almeida, an OpenAI researcher who helped invent RLHF, left the company to start Typesafe AI and released Jev, a non-LLM transformer model that outputs probabilities instead of text for software automation tasks. Jev is significantly faster and cheaper than existing LLMs while eliminating hallucinations, with developers reporting 5-20x performance improvements in classification and routing applications.
This paper addresses fundamental gaps in agentic software engineering: the requirement gap (requirements approximate stakeholder intent) and model gap (deployment models approximate real environments). The authors propose an assurance-revision loop using deployment evidence to iteratively narrow these gaps, framing the problem as resource allocation between human judgment and agent capability.
A developer created a tool called 'reroll' that generates LLM responses five times to test consistency and analyze variation patterns. The tool uses Haiku to categorize differences—whether answers are uniform, mostly uniform, split, or highly divergent—and identifies whether variations stem from different interpretations, different next moves, or contradictions that suggest hallucinations.
Axiom is a hybrid search engine combining BM25 and semantic vector search, designed to provide accurate, deterministic results without relying on cloud services. It operates locally for offline research, AI agent grounding, and sensitive data indexing, addressing concerns about AI hallucinations and data privacy in modern search.
A lawsuit accuses an AI security company of publishing hallucinated findings. The article appears to be a news digest covering multiple technology stories including cybersecurity threats, AI industry developments, and software updates.
PlanLint is an open-source AEC compliance checker that uses LLMs only for entity recognition and text parsing, while delegating all compliance verdicts to deterministic Python logic grounded in actual CAD geometry and codebook rules. The system models compliance as a bipartite graph in Neo4j, ingesting architectural drawings through specialized pipelines for different sheet types and applying rule-based verification to prevent AI hallucinations that could pass non-compliant designs.
Researchers demonstrated that prompt steering can reduce DeepSeek v4 Pro's hallucination rate below GPT and Claude by tailoring system prompts to the model's specific tendencies. Using the Omniscience benchmark, they found that custom mitigations focused on encouraging abstention significantly outperformed generic anti-hallucination instructions while maintaining accuracy.
A demonstration of how AI agents behave when given read-write access to their own context window. The agent self-corrects false notes, edits its own hallucinations, verifies its reasoning by querying copies of itself, and can be manipulated through context editing—sometimes retracting true statements when context is hidden or falsified.
AI systems make autonomous decisions at runtime without pre-approval, creating governance gaps that traditional oversight cannot address. Companies suffer real liability when AI models hallucinate or misclassify without human review, representing governance failures rather than technology failures. Effective risk management requires robust oversight structures, since no model will achieve zero error rates.
A developer created a tool that verifies AI-generated answers before displaying them to users, aiming to catch errors, hallucinations, and code problems before they're seen. The tool performs pre-execution verification to improve reliability of AI responses.
Aqqai is a tool that verifies AI-generated answers before displaying them to users, intercepting errors, hallucinations, and code anomalies at execution time.