Lightning Talks
Lightningat Understanding Higher-Level Intelligence from AI, Psychology, and Neuroscience Perspectives · 2024-06-24..06-28·
summary
Short study defines a five-level hierarchical scale for algorithm understanding and runs a five-question survey comparing students to GPT-3.5/4/4o on two algorithms: Euclidean vs Ford‑Fulkerson. Results show GPT‑4 equals PhD students on GCD/code tasks but fails on deeper mathematical reasoning and on constructing a graph needing six augmenting paths. A separate line formalizes multimodal interactions as redundancy/uniqueness/synergy, defines synergy via a max‑entropy coupling of unimodal marginals, and uses a mixture‑of‑experts to boost synergy‑heavy cases like sarcasm.
AI-written summary · about
chapters (13)
- 0:00Algorithm understanding in LLMs
- 2:10Can computers understand algorithms?
- 4:58Hierarchical scale of understanding
- 8:32Student vs LLM algorithm performance
- 12:40Q&A on algorithm understanding
- 19:47Multisensory AI principles
- 25:04Mathematical multimodal interactions
- 33:58Training for redundancy and uniqueness
- 38:30Neuroscience questions in AI
- 41:39NEMO model applications
- 50:32Parsing in NEMO
- 59:47NEMO research implications
- 1:00:08Q&A on NEMO model
related talks
25mWhy LLMs Still Fail: Reasoning, Robustness, and Safety
46mConversations with ChatGPT: illusion or reality?
01h 22mDebate: Sparks versus embers
01h 01mKeynote: Transformers: The Rise of ChatGPT
24mHarnessing the Power of LLMs in Advanced Analytics
01h 05mCan you trust your (large language) model?
42mQuestions for Theory in the New Age of Machine Learning
54mKeynote: Can you trust your (large language) model?