Using Large Language Models in the DoD Context

Dive deep into the world of Large Language Models (LLMs) with a special focus on their relevance and application within the Department of Defense (DoD). This comprehensive video serves as an essential primer for DoD personnel, shedding light on the forefront of AI technology, its potential uses, and the critical guidelines for its application within defense mechanisms.

Highlights:

• DoD Guidelines & Cautions: Learn about the interim guidance issued by the Chief Digital and Artificial Intelligence Office (CDAO) for LLM usage within DoD activities and why adherence is crucial.
• ChatBot Arena Insights: Discover which LLMs are leading the way in innovation and how they're ranked in the competitive landscape.
• LLM Training & Inference: Gain insight into the process of training large language models, including the resources and technologies involved.
• Intrinsic Nature of LLMs: Language “understanding”, empirical nature, emergent properties, in context learning, hallucinations, elementary prompt engineering to include chain of thought.
• Beyond LLMs: Explore other transformative AI technologies like AlphaFold and their implications for the future of DoD operations.
• Practical Demonstrations: Follow along with hands-on demonstrations on accessing and interacting with top-ranking models for both personal and official use, adhering to DoD guidance.
• Future Outlook: Peek into what the future holds for DoD's engagement with these advanced AI models and the strategic advantages they offer.

Topic Index:

• 01:08 CDAO Interim Guidance & LLM Cautions
• 02:23 Chatbot Arena Leaderboard
• 04:26 Overview of LLM Training & Resources
• 08:48 Training LLMs to Predicting the Next Word
• 11:08 Do These Models Really Understand?
• 12:45 Lack of Theory; Product of Trial & Error
• 13:18 Empirical and Strange Nature of LLMs
• 14:47 Hallucinations
• 17:05 Chain of Thought (CoT) Prompting
• 17:47 Suggested ChatGPT Custom Instructions
• 18:58 Bigger is Better and Emergent Properties
• 20:06 Its Not Just Size that Matters :-)
• 20:49 In Context Learning
• 22:53 Voyager: LLMs are Problem Solvers
• 23:54 AI Agents and AutoGen Framework
• 24:32 Custom GPTs; Tailoring LLMs for Specific Use
• 25:26 Tool Users; Glimpses of Problem Solvers
• 26:01 Demo Tool Use, Problem Solving and CoT
• 27:12 LLMs are Quickly Becoming Multi-Modal
• 27:35 Transformers Good for More Than LLMs
• 28:22 Multi-Modal LLMs; Early & Late Fusion
• 30:23 More than Chatbots; Are LLMs Future CPUs?
• 31:15 AI & Model Based System Engineering
• 33:32 Using AI to make Emirates Racing Team fly!

Please first watch these videos:

• Insights into the Fundamentals of a Simple Neural Network; https://media.dau.edu/media/t/1_m0aa45bw
• How AI Learns to Talk; https://media.dau.edu/media/t/1_vje3ghev/310800832
• How ChatGPT Works in Inference; https://media.dau.edu/media/t/1_8bcxvf7y

The link to a course playlist of DAU recommended AI courses: https://dau.csod.com/ui/lms-learner-playlist/PlaylistDetails?playlistId=00118adb-20e1-4dc5-95a8-9ffd03ab7f70

You will need a DAU account to access these resources. If you are a DoD member and need a DAU account and you can request one here: https://www.dau.edu/faq/p/New-DAU-Account

Other promised links:

• ChatBot Arena: https://huggingface.co/spaces/lmsys/chatbot-arena-leaderboard
• Andrej Karpathy’s “[1hr Talk] Intro to Large Language Models”: https://www.youtube.com/watch?v=zjkBMFhNj_g&t=2352s
• Barron Stone’s “There is No Spoon: U.S. Air Force Digital Acquisition Strategy (Summary)”: https://www.youtube.com/watch?v=dEcPlqImjWc&t=7s

Additional References used for creating this video:

• Llama 2: Open Foundation and Fine-Tuned Chat Models; Touvron et al; https://arxiv.org/pdf/2307.09288.pdf
• Geoffrey Hinton and Andrew NG - Does Ai Understand - AGI; https://youtu.be/6-a33BI6fnk?si=Kua7XRDxo9G_0VL8
• The Godfather in Conversation: Why Geoffrey Hinton is worried about the future of AI; https://www.youtube.com/watch?v=-9cW4Gcn5WY&t=1688s
• The Reversal Curse: LLMs trained on "A is B" fail to learn "B is A“; Berglund et al; https://arxiv.org/pdf/2309.12288.pdf
• Survey of Hallucination in Natural Language Generation; Ziwei Ji et al; https://arxiv.org/pdf/2202.03629.pdf
• SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection…; Manakul et al; https://arxiv.org/pdf/2303.08896.pdf
• Large Language Models are Zero-Shot Reasoners; Kojima et al; https://arxiv.org/pdf/2205.11916.pdf
• Training Compute-Optimal Large Language Models; Hoffmann et al; https://arxiv.org/abs/2203.15556
• GPT 4 Technical Report; OpenAI; https://arxiv.org/pdf/2303.08774.pdf
• Emergent Abilities of Large Language Models; Wei et al; https://arxiv.org/pdf/2206.07682.pdf
• Language Models are Few-Shot Learners; Brown et al; https://arxiv.org/pdf/2005.14165.pdf
• Voyager: An Open-Ended Embodied Agent with Large Language Models; Wang et al; https://arxiv.org/abs/2305.16291
• AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation; Wu et al; https://arxiv.org/pdf/2308.08155.pdf
• Multimodal Learning with Transformers - A Survey; Xu et al; https://arxiv.org/pdf/2206.06488.pdf
• Small Language Model Meets with Reinforced Vision Vocabulary; Wei et al; https://arxiv.org/pdf/2401.12503.pdf
• Vary-toy demo; Wei et al; https://varytoy.github.io/
• The capabilities of multimodal AI | Gemini Demo; https://youtu.be/UIZAiXYceBI?si=dGZcn2YlUlZeqcjv
• America’s Cup Sailing: Using AI to make Emirates Team New Zealand fly; McKinsey & Company; https://www.youtube.com/watch?v=uXihkPI-LyM&list=LL&index=133
• Dr. Will Roper | The Urgent Need for The Pentagon to Unlock AI’s Potential; https://youtu.be/MadVS_IE0KM?si=kVjyneK1WeiPRPlX