Do we know where we are going to?

From script by Lou Cripps

As infrastructure organisations start to make use of large language models (LLM) – popularly labelled artificial intelligent, though they are not actually thinking – do we know where we want them to take us?

Talking Infrastructure members are involved in some experimentation here. Blue Mountains City Council, for example, is trialling such technology to automate rural road inspections from LLM interpretation of videos of defects.

Lou Cripps is training models with curated Asset Management information. That is, instead of letting something like ChatGPT loose on all sources, weighting scripts to focus on known and reliable material like SAM newsletter articles. (Otherwise, they are biased towards financial AM, since there’s much more of that material around.)

As Talking Infrastructure looks at how such tools can help make good AM practice more accessible, and what principles we need when, not if, organisations try using LLM in decision processes such as where to schedule road maintenance:

What have you used so far? Where do you think we can make best use of LLM?

One Thought on “Do we know where we are going to?

  1. Lou Cripps on March 26, 2025 at 12:10 am said:

    Just to clarify, it’s not exactly “training” in the machine learning / AI sense (which involves adjusting model weights based on new data). What we are doing is more like context provisioning—supplying relevant, trusted information to guide the model’s outputs.

    Other terms that might fit include:

    Grounding – Using external facts or sources to make answers more accurate and contextually relevant.

    Knowledge injection – Temporarily adding external data or documents to improve task-specific output.

    Resource augmentation – Supplementing existing knowledge with focused material to enhance accuracy and relevance.

    Context provisioning – Providing specific background information to steer the outcome of a task or conversation.

    “Grounding” and “context provisioning” are probably the most widely used terms I have come across in AI blogs.

    I am NOT an AI expert, just someone grappling with how to best deploy these tools to benefit asset management—especially focused on how to reduce risks by understanding the principles we will need to follow.

    I am curious to hear how others are approaching this.

Leave a Reply

Your email address will not be published. Required fields are marked *

Post Navigation