Create 2024-08-28.md

Adds meeting minutes/agenda SPDX AI Team Meeting 2024-08-28

ai/2024/2024-08-28.md

@@ -0,0 +1,59 @@
+# SPDX AI Team Meeting 2024-08-28
+
+## Attendees
+- Arun Prabhakar
+- Arthit Suriyawongkul
+- Gopi Krishnan Rajbahadur
+- Helen Oakley
+- Kate Stewart
+- Karen Bennet
+
+## Agenda   
+- 3.1 Strategy Release - Gopi
+- New Business
+- Other Business
+- Conferences
+  - SBOM a RAMA: Sept 11 - 13 https://www.cisa.gov/news-events/events/sbom-rama-fall-2024
+
+## Notes
+- Gopi presented on additional fields needed for 3.1 on GenAI
+- multimodal world is in the future; new generation software
+- tradiational programderstand what agents will have control as well as how storage is used- What
+- interaction design interface, agents and protocol decide how to solve, need to training with architecture, how prompts wares and agency ware.
+- need a business workflow (in the operation profile?)
+- How to test this new architecture - very hard, it's not about hosting one model,  many models / how to trace the whole process
+- software around the model (has it's  own development process with RAG, etc, and devops is  need cache, batch, workflow, how to monitor, feedback mechanism), to monitor, log - many new steps involved
+- what does this mean for data, lable or unstructured data used for training
+- data is more now, no code now, LLM makes the decision, so data is the 'new' code
+- change is that humans aren't in the loop, it's machine intelligence
+- data alighment - right data is selected
+- model data - how we teach, optimization
+- let's start with prompts for 3.1 -
+- from Arun: I believe this is where Multi-Agent Framework comes 🙂 … One to build and others to test, provide security, etc
+- prompts to instruct foundation models (primary communicate channel for our AI CPU)
+- chatgpt (underlying prompt); then user prompts, need a system prompt now 
+- what are the components to a prompt : text related, context related. there are many components for a prompt.  like batch statements
+- chain of prompting is an examples, need to capture patterns of prompts
+- ask chatgrp on the fields (prompt_id), licnesing, input context, modification / version h
+istory
+- EU AI act includes intelligent based solutions is that software 3 or 4
+ -What is a AI BOM to focus on In Gopi's opinion, EU AI act is very confusing.  AI BOM should be focusing on software level (complete transparency).  The scope of AI BOM is about how model is being used.   is AI BOM different than  SBOM.  
+ -Arum suggested defining scope for 3.0 (prompt); 3.1  (AI Assistants and AI Agents (instead of saying Generative AI)
+ - can't focus on RAG and GenAI in general
+ - scope use case : LLM/Agent Integrated Apps"
+ - Gopi:  FM Ware is the use case for 3.1 
+ - Using definition of software 3.0 and 4.0; but open to other terms.  Arun suggested using Gartner glossary
+ - AI BOM is for AI deployers and makers
+ - The terminology has been accepted  but if there is an alternative  suggestions, please bring forward
+ - Prompts (Rag, database, etc) fields - to be discussed later
+ - AI BOM Workshop: https://www.aiwarebootcamp.com/
+ - Other relationship: https://w3c.github.io/dpv/2.0/dpv/modules/risk.html
+ - Paper to <harm> review: https://www.aiaaic.org/projects/ai-algorithmic-risks-harms-taxonomy
+- Next week's meeting: Harm categoriesming (human writing code) 2nd generation (traditional AI model, existing AI BOM goals), 3rd gneeration (genAI) (FM Ware);  LLM = foundational models
+- future has smaller models, so referring to foundational models (programming via promtp)
+- we have agents that can 3.0, 3.1, 4.0 agents now with these foundational models
+- new models of mind ware - specific what you want and agents figures out a path to get a solution
+- these agents will have reasoning capability
+- AI ware, foundation models are the newparadyms for the BOM's future
+- need transparency through the development process
+- BOM need to un