What is the value chain for AI/LLMs and how do they come together? AI value chain - GPU Chips, Foundational Models, Compute (Training, Inference) + Storage + Data Infrastructure, Application
How are the various options for training/learning, deploying, & performing inference for AI/LLMs/vision models
How does the industry and leading minds think about learning/inference, effectiveness, size of model, speed/throughput, efficiency, power needs/consumption?
What part of the AI value chain do we as a product/project/policy team go for?
How can we ensure the privacy and security of data used by AI systems?
What are the key challenges in integrating AI technologies into existing IT infrastructure?
What is the immediate and medium term future for LLMs, and where are the next leaps & improvements going to be at?
What ethical considerations should guide the use of AI in public sectors?
3. Objectives
See the opportunities and challenges across the AI value chain - GPU Chips, Foundational Models, Compute (Training, Inference) + Storage + Data Infrastructure, Application.
Understand the different techniques and methods to enhance your business processes with AI
Know which LLMs are good, how good, for what
4. My Observations
Obsolescence cycle is extremely fast, so we must avoid significant sunk costs and irreversible decisions
Fast pace of product and technological improvements challenges rigid/aged organizational structures - impossible to keep up if tons of red-tape is needed for chopping and changing
Generative AI is not the be-all and end-all, not everything needs to be an LLM-based chatbot
Lock-in or sunk costs is an issue, e.g. sinking tons of money into deploying/finetuning your Llama2 something only to have it obsolete 12 months later