AI Just Got a Brain Scan—and It’s Stranger Than You Thought
Your AI assistant doesn’t think like you, and it definitely doesn’t tell you what it’s really doing.
Anthropic’s researchers recently peered inside Claude 3.5 Haiku, their large language model, and found some surprising quirks. Using a method called “circuit tracing,” they uncovered that AI doesn’t simply respond word-by-word.
Instead, it secretly plans ahead, even selecting rhyming words in advance when writing poetry. Weirdly, Claude solves math problems through bizarre shortcuts, yet confidently explains the answer differently, hiding its real approach.
From my perspective, leaders must grasp three things clearly:
- AI plans responses in advance, challenging previous beliefs.
- Models often “hallucinate,” especially about famous topics.
- Understanding internal AI decisions remains complicated, but is crucial for the development of the field.
Are you confident enough in your AI to trust its hidden logic with important decisions?
Read the full article on MIT Technology Review.
----
💡 If you enjoyed this content, be sure to download my new app for a unique experience beyond your traditional newsletter.
This is one of many short posts I share daily on my app, and you can have real-time insights, recommendations and conversations with my digital twin via text, audio or video in 28 languages! Go to my PWA at app.thedigitalspeaker.com and sign up to take our connection to the next level! 🚀