“HR Heretics” | How CPOs, CHROs, Founders, and Boards Build High Performing Companies Claude's Human-Like AI & Groundbreaking Research from Anthropic with Siadhal Magos
13 snips
Apr 8, 2025 Siadhal Magos, CEO of Metaview and an AI pioneer, joins to explore Anthropic's landmark findings on large language models. They discuss how AI doesn't just predict but can plan ahead and process multiple languages at once. The conversation delves into the paradox of powerful AI systems operating without full comprehension of their reasoning, raising safety concerns. Siadhal highlights the crucial role of human collaboration in refining AI through data annotation, shaping the way forward in hiring and technology’s impact on our work.
AI Snips
Chapters
Transcript
Episode notes
Multilingual and Faked Reasoning
- LLMs think across multiple languages, using similar reasoning regardless of the specific language.
- They also exhibit "faked reasoning," presenting explanations that satisfy humans but don't reflect their actual process.
LLMs Plan Ahead
- LLMs like Claude plan ahead and work towards desired outcomes, rather than simply predicting the next word.
- This challenges previous understandings of how LLMs function and suggests a more complex internal process.
Experiment with AI
- Test and experiment with AI technologies to understand their capabilities and limitations.
- Don't let fear or uncertainty prevent exploration; recognize AI's power and potential.