Probably, a startup that raised $9 million in seed funding, is developing a data science tool to combat hallucinations and factual errors in LLMs. Their "data science mech suit" system uses a deterministic validator to check LLM outputs, allowing for smaller, more efficient models and reduced costs.
KPMG has withdrawn its global report on "Agentic AI" after major companies named in it complained that their supposed achievements were fabricated. The report contained false claims and bogus case studies, allegedly generated by AI "hallucinations" that KPMG staff failed to verify.
KPMG's report on AI, "Total Experience: Redefining Excellence in the Age of Agentic AI," contained numerous AI hallucinations, including fake citations and inaccurate claims about companies like Emirates, UBS, and Swiss Federal Railways. GPTZero, an AI content detection tool maker, found that most citations were inaccurate or fabricated, and about half of the paper's claims were false. KPMG has since withdrawn the report and is investigating.
Experts caution against relying solely on AI for financial advice due to risks of inaccuracy, bias, and privacy concerns. While AI can be a learning tool, cross-checking information and using regulated sources are crucial.
Attorneys admitted to a federal court that they submitted fabricated quotations generated by AI in a lawsuit against the Trump administration over federal layoffs. The mistake occurred when using Anthropic's Claude Console, leading to "phantom quotations."
OpenAI's newest default model for ChatGPT might not make stuff up as much. Hallucinations have been an ongoing problem for AI models, but OpenAI says its new GPT-5.5 Instant model has "significant improvements in factuality across the board." The company claims that, based on "internal evaluations," GPT-5.5 Instant produced "52.5% fewer hallucinated claims" than its […]
Sullivan & Cromwell, one of America's largest law firms, has apologized to a federal bankruptcy judge for AI-generated errors in a court filing. The firm submitted a letter on April 18 to the US Bankruptcy Court for the Southern District of New York, admitting that an emergency motion in the Prince Global Holdings bankruptcy case contained fictitious case names, fabricated quotes, and invented statutory provisions. Andrew Dietderich, founder of S&C's restructuring group, apologized on behalf of the firm, stating they had not followed protocols in preparing the document. The firm, which holds an enterprise license for ChatGPT, noted it instructs lawyers to 'trust nothing and verify everything' and that failure to verify AI output violates firm policy.
DÉCRYPTAGE - Ce rapport de référence, récemment publié par l’université de Stanford (Californie), analyse les tendances et les performances de l’intelligence artificielle à travers le monde.
Sullivan & Cromwell, a top Wall Street law firm, has apologized to a New York federal judge for AI-generated errors in a court filing. The firm admitted to misquoting the US bankruptcy code and incorrectly citing cases in a filing made on April 9. The errors were uncovered by rival firm Boies Schiller Flexner. The case involves liquidators from the British Virgin Islands pursuing actions against Prince Group, owned by Chinese-born businessman Chen Zhi, who faces US charges of wire fraud and money laundering related to alleged forced-labour scam compounds in Cambodia.
The rise of AI has brought an avalanche of new terms and slang. Here is a glossary with definitions of some of the most important words and phrases you might encounter.
