Back to News

Google's healthcare AI made up a body part

Hacker News - AI
Aug 11, 2025 22:35
workfromspace
1 views
hackernewsaidiscussion

Summary

Google's healthcare AI, Med-Gemini, mistakenly referenced a non-existent body part, "basilar ganglia," in a research paper, highlighting ongoing issues with AI-generated hallucinations. This incident underscores the need for improved accuracy and reliability in medical AI systems, especially when used in critical healthcare contexts.

Article URL: https://www.theverge.com/health/718049/google-med-gemini-basilar-ganglia-paper-typo-hallucination Comments URL: https://news.ycombinator.com/item?id=44870249 Points: 2 # Comments: 0
Read Full ArticleMore News

Related Articles

AI Summarizer: Web, YouTube and PDFs with ChatGPT, Claude and More

Hacker News - AIAug 12

The article highlights the "AI Summarizer" Chrome extension, which leverages advanced AI models like ChatGPT and Claude to generate concise summaries of web pages, YouTube videos, and PDFs. This tool demonstrates the growing accessibility and practical integration of AI-powered summarization into everyday browsing, streamlining information consumption and enhancing productivity. Its development reflects broader trends in applying generative AI to improve user experience and information management.

The Quiet Disappearance of Skeptics in the AI Gold Rush

Hacker News - AIAug 11

The article discusses how vocal skeptics and "AI doomers" have become less prominent amid the rapid growth and optimism in the AI industry. It suggests that the prevailing enthusiasm may be overshadowing critical debate about AI's risks and long-term societal impacts, raising concerns about the lack of balanced discourse in the field.

Show HN: Persistent Mind Model – Portable AI Personas for Any LLM

Hacker News - AIAug 11

The Persistent Mind Model enables users to create and evolve personalized AI personas that retain memories and behaviors across different LLMs (like GPT, Claude, LLaMA, Mistral) and devices, overcoming the usual limitations of AI “personalities” tied to specific platforms. This portable, vendor-independent approach allows for long-term, user-driven AI development, potentially advancing personalization and interoperability in the AI field.