Machine Learning Street Talk (MLST) Dr. Brandon Rohrer - Robotics, Creativity and Intelligence
25 snips
Feb 13, 2024 Brandon Rohrer, a seasoned data science leader with a Ph.D. from MIT, dives deep into the fascinating world of machine learning and robotics. He discusses the nuances of Reinforcement Learning from Human Feedback, comparing it to traditional coding tasks. The conversation explores the creativity of transformer models and the philosophical dimensions of machine intelligence. Rohrer emphasizes the challenge of training models in dynamic environments and advocates for biologically inspired methods, all while critiquing our anthropocentric definitions of intelligence and creativity.
AI Snips
Chapters
Books
Transcript
Episode notes
LLM Limitations
- LLMs excel at regurgitating familiar patterns, mimicking intelligence without true understanding.
- This raises concerns about their ability to reason in novel situations, despite convincing answers.
LLM Coding Examples
- Brandon Rohrer shares examples of LLMs generating code.
- While helpful for boilerplate, they can introduce subtle bugs in novel scenarios due to a lack of true understanding.
GPT as Yardstick
- Use GPT as a yardstick for task complexity: if it's helpful, the task might lack intellectual challenge.
- Consider changing your profession if GPT easily automates your innovative work.
