Thomas Hirschmann's Highlights on 'Large language models can do jaw-dropping things. But nobody knows exactly why.' | Glasp