🇨🇦Canadian Tribunal Rejects Fabricated Case Law In Canada a family couple relied on Microsoft Copilot to generate legal precedents in a condo dispute—only to discover that nine out of ten cited rulings didn’t exist. The Civil Resolution Tribunal found the cases to be AI “hallucinations,” raising serious concerns about the reliability of AI-generated legal research. While AI can streamline legal work, this case underscores a fundamental risk: without proper verification, reliance on AI-generated case law can undermine legal arguments and credibility. . #AI#LegalTech#AIEthics#Hallucinations
TGTGInsighttelegram intelligenceLIVE / telegram public index
DN42 access 本服务为那些无法轻松访问自身网络的用户以及希望体验 dn42 但又不想承担维护自有网络成本的用户提供 dn42 连接 默认情况下，地址从/96地址块中分配，如果您希望租用独立的/96前缀或更大的地址空间，请按照联系方式联系我 所有公开的PoP均已屏蔽来自中国境内的 IP 地址。如果您确实需要dn42 access，请与我联系并提供合理的理由 该服务由AS4242423377提供 - - - - - - - The service provides DN42 connectivity to members who cannot easily access their own networks, as well as to those who would like to explore DN42 without the overhead of maintaining their own network. By default, addresses are allocated from a /96 block. If you wish to lease a dedicated /96 prefix or a larger address space, please contact me using the methods provided in the contact information. All publicly accessible PoP are blocked for IPs originating from within China. DN42 access from within China is not publicly available. If you genuinely require access, please contact me and provide a valid justification. Hosted by AS4242423377. Policy 本服务需要花费时间和金钱才能运行，但为了您的利益，我们免费提供。使用本服务是一种特权，而非权利。您必须合理使用本服务，以确保其他用户也能继续享受同样的便利。任何滥用、误用或干扰服务或其他用户的行为都可能导致您的访问权限立即被暂停或终止。 滥用行为包括但不限于： - 过度使用资源 - 黑客攻击、病毒、木马等，或任何其他可能损害服务或对服务及其用户造成风险的干扰行为 - 传播可能导致民事或刑事责任的不良内容 - - - - - - - This service require real time and financial resources to operate, yet are provided free of charge for your benefit. Access to the services is a privilege, not a right. You must use the services responsibly and considerately to ensure that other users can continue to enjoy the same opportunities. Any misuse, abuse, or activities that disrupt the service or other users may result in immediate suspension or termination of access. Abuse could include, but is not limited to: - Excessive use of resources - Hacking, viruses, trojans etc or any other disruption that could harm or create risk to the services or its users - Distribution of objectional content that could create a civil or criminal liability PoP ## Toronto, Canada Prefix: fdb6:fc6a:e66c:724f:fad1:d2cf::/96 Zerotier: 4753cf475f65b0fb ## Los Angeles, USA coming soon #announcement#service
Hashtags
Results
找到 4 条相似帖子
搜索 #hallucinations
当前筛选 #hallucinations清除筛选
🇿🇦South Africa Withdraws AI Policy Over Hallucinated Sources South Africa has withdrawn its draft national AI policy after discovering that at least 6 of its 67 academic citations were AI-generated and referred to non-existent journal articles. Communications Minister Solly Malatsi stated that the most plausible explanation is the inclusion of unverified AI-generated references, calling the lapse a failure that “compromised the integrity and credibility” of the policy. The draft policy had proposed establishing a national AI commission, an AI ethics board, and a regulatory authority, alongside incentives such as tax breaks and grants to support AI infrastructure. The issue was identified after News24 found fabricated citations, later confirmed by journal editors. The policy will be revised before being reissued, and the minister indicated there would be consequences for those responsible. The case highlights risks of using generative AI in policy drafting without verification. A Nature study cited in the report found that over 2.5% of academic papers in 2025 contained at least one potentially hallucinated reference, up from 0.3% in 2024, amounting to more than 110,000 papers. #AIRegulation#AIethics#Hallucinations#PublicPolicy#AIGovernance
Google Facts™ [ ️@googlefactss🌎]
@googlefactss · Post #40980 · 2026/04/27 05:23
People who eat the mushroom Lanmaoa asiatica raw or undercooked have reported seeing tiny human-like figures moving around them. These are called lilliputian hallucinations. Reports from people across different cultures and backgrounds describe similar details, including small figures walking on floors and furniture. The effects can begin 12–24 hours after eating and may last 1–3 days. Some cases are serious and require hospital care. Don't try this yourself.. 🍄😵💫🧚♀🦄🍄🟫 [Read more 1] [Read more 2] [Read more 3] @googlefactss #Mushrooms#ScienceFacts#Hallucinations#Nature#DidYouKnow If you have ideas or feedback contact us: @Googlefactss_Feedback_bot
Machinelearning
@ai_machinelearning_big_data · Post #8518 · 2025/09/11 17:11
🔥WFGY 2.0 — Semantic Reasoning Engine for LLMs (MIT) Это движок с открытым исходным кодом, цель которого — уменьшить галлюцинации и логические сбои в системах типа RAG / LLM, особенно когда: - источники OCR-текста плохо распознаются, - происходит «semantic drift» (когда ответ уходит от вопроса), - «ghost matches», когда извлечённый фрагмент кажется релевантным, но на самом деле нет. Обычно ошибки ловят уже в готовом сгенерированном тексте, из-за чего они часто повторяются. В Semantic Reasoning Engine всё наоборот: если система видит, что рассуждения «кривые» или сбились с курса, она останавливается, сбрасывается или ищет другой путь и отвечает только когда состояние стабильно. 🛡Авторы называют это semantic firewall - семантический «файрвол». Проверки встроены прямо в процесс мышления модели, а не поверх ответа с фильтрами или регексами. Это помогает избегать ошибок до того, как они попадут в вывод. 📌 Проект включает карту из 16 типичных ошибок LLM: - неверный поиск данных, - сбившаяся логика, - «провалы памяти», - путаница ролей агентов и другие. Для каждой есть простое текстовое исправление. Никаких SDK — достаточно вставить инструкции прямо в промпт. 🟢Как модель решает, правильные ли ответ генерируется: - ΔS (drift) - не уходит ли смысл слишком далеко от шага к шагу - λ (convergence) - сходится ли рассуждение к решению или крутится в цикле - Coverage — достаточно ли фактов и аргументов учтено Если все три условия выполнены, ответ считается «качественным». 🟢В тестах стабильность вывода выросла до 90–95% против обычных 70–85% у традиционных подходов. ▪Github: https://github.com/onestardao/WFGY @ai_machinelearning_big_data #ai#llm#opensource#reasoning#hallucinations#promptengineering