"The Cognitive Revolution" | AI Builders, Researchers, and Live Player Analysis

E14: The Reasoning Revolution with Ought's Jungwon Byun and Andreas Stuhlmüller

19 snips
Apr 6, 2023
Discover the transformative potential of AI tools like Elicit in research methodologies. The co-founders of Ought discuss the limitations of current AI models in reasoning and the need for transparency in decision-making. Dive into errors in human-AI collaboration and the importance of context in interpreting AI outputs. They also tackle the evolution of machine learning, emphasizing user feedback and innovative workflows to enhance research efficiency. This conversation explores how AI could bridge disagreements and improve trust in research.
Ask episode
AI Snips
Chapters
Transcript
Episode notes
INSIGHT

ML's Limitations in Reasoning

  • Machine learning excels at imitation and reward optimization, like game playing.
  • Reasoning and decision-making lack clear objectives for ML, hindering its advancement.
INSIGHT

Reinforcement Learning's Limits

  • AI's strength in reinforcement learning relies on well-defined tasks and objectives.
  • Open-ended reasoning lacks these, leading to a focus on persuasive outputs over true reasoning.
ANECDOTE

Early Ought Experiments

  • Ought's early research used human stand-ins for AI, exploring task decomposition.
  • Surprisingly, current AI systems resemble these humans more than expected.
Get the Snipd Podcast app to discover more snips from this episode
Get the app