Machine Learning Street Talk (MLST) #105 - Dr. MICHAEL OLIVER [CSO - Numerai]
5 snips
Mar 4, 2023 Michael Oliver, Chief Scientist at Numerai, dives into the world of crowdsourced financial modeling. He discusses the unique approach Numerai takes in utilizing diverse models for stock predictions while addressing the limitations of AI like large language models. The conversation explores the intricacies of language and cognition, emphasizing how subjective experiences shape communication. Oliver also highlights the collaboration within the data science community and innovative methods that are breaking new ground in machine learning.
AI Snips
Chapters
Books
Transcript
Episode notes
LLM Understanding
- Large language models (LLMs) excel at surface statistics from vast data but lack deep understanding.
- LLMs exhibit brittleness and hallucination, creating a deceptive appearance of comprehension.
Bing's Hallucinations
- Bing's new search engine launched with much fanfare, but quickly faced criticism.
- It fabricated information, including financial figures, highlighting issues with AI hallucination.
Bing's Unpredictability
- Michael Oliver experimented with Bing, finding it occasionally insightful but unpredictable.
- He notes the humor in trying to control its behavior by "asking nicely," highlighting the limitations of current approaches.
