🚀🔍 AI papers deep dive: LLM understanding, RAG, CoT

🌐 RA-DIT Retrieval-Augmented Dual Instruction Tuning: Enhancing RAG systems by fine-tuning both the LM & retriever. Implementation in LlamaIndex.
📆 FreshLLM: Leveraging search engines to keep LLMs updated with the latest info.
🧠 Analogical Reasoners: Automatically guides the reasoning of LLMs for a better chain-of-thought.
✨ On AI transparency & interpretability, two pivotal papers:
– Representation Engineering: A Top-Down Approach to AI Transparency
– Decomposing Language Models Into Understandable Components
#ArtificialIntelligence #AIpapers

Links:
https://arxiv.org/abs/2310.01352
https://twitter.com/jerryjliu0/status/1709646787076935818
https://arxiv.org/pdf/2310.03214v1.pdf
https://arxiv.org/pdf/2310.01714.pdf
https://www.ai-transparency.org/
https://www.anthropic.com/index/decomposing-language-models-into-understandable-components
https://transformer-circuits.pub/2023/monosemantic-features/index.html


00:00 intro
01:04 RA-DIT Retrieval-Augmented Dual Instruction Tuning
04:17 FreshLLMs
07:02 Analogical Reasoners
10:07 Representation Engineering
14:18 Decomposing Language Models Into Understandable Components


🔔 SUBSCRIBE to my channel: https://www.youtube.com/c/SophiaYangDS?sub_confirmation=1

⭐ Stay in touch ⭐
📚 DS/ML Book Club: http://dsbookclub.github.io/
▶ YouTube: https://youtube.com/SophiaYangDS
✍️ Medium: https://sophiamyang.medium.com
🐦 Twitter: https://twitter.com/sophiamyang
🤝 Linkedin: https://www.linkedin.com/in/sophiamyang/
💚 #ai