Applying LLMs
Key takeaways
Creating value
LLMs as components
The idea is to move from ‘word models’, which describe problems and situations, to ‘world models’, which formalise what the word model implies to enable inferences to be made. The word to world concept is to use a code generating LLM to generate logic programmes that can then be integrated with an applications knowledge base for inference.
LLMs in application context
Whilst infrastructures cannot guarantee application success, they can provide a mechanism that both eliminates the drivers of failure and, at the same time, eliminates many proposed applications that are doomed from the start. This is very important for any organisation trying to exploit machine learning, since poorly founded projects can absorb so much investment that they derail the whole attempt.
