Published Jul 29, 2024

Prof. Subbarao Kambhampati - LLMs don't reason, they memorize (ICML2024 2/13)

Prof. Subbarao Kambhampati discusses the limitations of large language models, emphasizing their tendency to memorize rather than reason. He proposes the LLM Modulo Framework that enhances reasoning by integrating LLMs with external systems, combining creativity with logical verification for greater reliability in complex tasks.

Episode Highlights

Topics covered

Episode Highlights

Related Episodes

Subbarao Kambhampati - Do o1 models search?
Answers 383 questions
Jürgen Schmidhuber - Neural and Non-Neural AI, Reasoning, Transformers, and LSTMs
Answers 383 questions
#90 - Prof. DAVID CHALMERS - Consciousness in LLMs [Special Edition]
Answers 383 questions
Mahault Albarracin - Cognitive Science
Answers 383 questions
Prof. Murray Shanahan - Machines Don't Think Like Us
Answers 383 questions
#73 - YASAMAN RAZEGHI & Prof. SAMEER SINGH - NLP benchmarks
Answers 383 questions
"AI should NOT be regulated at all!" - Prof. Pedro Domingos
Answers 383 questions
MLST #78 - Prof. NOAM CHOMSKY (Special Edition)
Answers 383 questions
Explainability, Reasoning, Priors and GPT-3
Answers 383 questions
How Do AI Models Actually Think? - Laura Ruis
Answers 383 questions
#043 Prof J. Mark Bishop - Artificial Intelligence Is Stupid and Causal Reasoning won't fix it.
Answers 383 questions
#57 - Prof. Melanie Mitchell - Why AI is harder than we think
Answers 383 questions
NLP is not NLU and GPT-3 - Walid Saba
Answers 383 questions
#96 Prof. PEDRO DOMINGOS - There are no infinities, utility functions, neurosymbolic
Answers 383 questions
(Music Removed) #90 - Prof. DAVID CHALMERS - Consciousness in LLMs [Special Edition]
Answers 383 questions

Dexa/Machine Learning Street Talk (MLST)

Prof. Subbarao Kambhampati - LLMs don't reason, they memorize (ICML2024 2/13)

Topics covered

Popular Clips

Reasoning in AI

End-to-End Verification

Creativity and Knowledge

Understanding LLMs

Tools and Knowledge

LLMs in Planning

Language vs. Formality

Misconceptions in AI

Style vs. Correctness

LLMs and Reasoning

Emergent Reasoning Insights

Logical Reasoning Challenges

Verification and Trust

Language Model Insights

Episode Highlights

LLM Modulo Framework

Creativity vs. Reasoning

Limitations of LLMs

Related Episodes

Subbarao Kambhampati - Do o1 models search?

Jürgen Schmidhuber - Neural and Non-Neural AI, Reasoning, Transformers, and LSTMs

#90 - Prof. DAVID CHALMERS - Consciousness in LLMs [Special Edition]

Mahault Albarracin - Cognitive Science

Prof. Murray Shanahan - Machines Don't Think Like Us

#73 - YASAMAN RAZEGHI & Prof. SAMEER SINGH - NLP benchmarks

"AI should NOT be regulated at all!" - Prof. Pedro Domingos

MLST #78 - Prof. NOAM CHOMSKY (Special Edition)

Explainability, Reasoning, Priors and GPT-3

How Do AI Models Actually Think? - Laura Ruis

#043 Prof J. Mark Bishop - Artificial Intelligence Is Stupid and Causal Reasoning won't fix it.

#57 - Prof. Melanie Mitchell - Why AI is harder than we think

NLP is not NLU and GPT-3 - Walid Saba

#96 Prof. PEDRO DOMINGOS - There are no infinities, utility functions, neurosymbolic

(Music Removed) #90 - Prof. DAVID CHALMERS - Consciousness in LLMs [Special Edition]

Prof. Subbarao Kambhampati - LLMs don't reason, they memorize (ICML2024 2/13)

Topics covered

Popular Clips

Episode Highlights

LLM Modulo Framework

Creativity vs. ReasoningSubbarao Kambhampati explores the creative capabilities of Large Language Models (LLMs) and their limitations in reasoning. He advocates for hybrid systems that combine LLMs with external verification to enhance their reliability and application across fields.

Creativity vs. Reasoning

Limitations of LLMsSubbarao Kambhampati critiques the reasoning capabilities of large language models (LLMs), highlighting their reliance on memorization over logical reasoning. He advocates for hybrid systems that incorporate external verification to ensure output accuracy.

Limitations of LLMs

Related Episodes