• Drifft Newsletter
  • Posts
  • Two trends you shouldn't ignore in AI – and how mining benefits

Two trends you shouldn't ignore in AI – and how mining benefits

How professional work is getting augmented with AI agents.

💡What’s New: The two trends making money in language model applications

🤔 Opinion: Augmenting the shrinking mining workforce (and other trends)

🛠️ Tools & Data: A collection of the latest, greatest opensource projects

Sponsor: Make your social profile stand out with a Bettershot AI headshot.

Enter code ‘DAVID’ at checkout

💡What’s new

In the past few newsletters I’ve defined modern AI as a set of “models” that find uses in many tasks like programming and mineral exploration. In this edition, we’ll look broadly at generating value by automating back-office tasks with these tools.

Demand for minerals in tech economies is growing rapidly, but so are challenges like

  • Inflation > higher costs eat up profits;

  • Most of the easily accessible ore is already claimed;

  • New environmental policies make it more costly to mine;

  • Implementing AI requires a big investment in hard-to-find talent.

Mining engineers must now explore and extract minerals from increasingly complex and remote sites. Additionally, there is growing pressure to minimize the environmental impact of mining operations. Mining engineers must develop innovative solutions to reduce water usage, energy consumption, and greenhouse gas emissions.

Mine owners, professionals, and startups working in this industry can improve their competitive positions by applying a few new techniques to begin automating back-office work.

In my last post I detailed how the professional work model is shifting. Andrew Ng, founder of deeplearning.ai, mentioned this in a recent X tweet–instead of using monolithic software packages to do work, professionals will use AI assistants called agents.

Automation of some white-collar work is going to be good news for a lot of businesses. It can lower costs to adopt advanced software automation in industries where added efficiencies translate into big big profits–like mining!

There are two important sub-themes to track in relation to AI agents:

  1. RAG – stands for Retrieve, Augment, Generate, a three-step reasoning pipeline for language models:

    • step 1: user asks a question or requests some information;

    • step 2: relevant chunks of data are fetched from a database (e.g., the web) using a similarity measure;

    • step 3: a language model synthesizes an answer to satisfy the request.

  2. Chain of Thought - a strategy for having a language model plan your work and subsequently work its own plan, step-by-step. Yes, it can!

By putting these two ideas together, it’s possible to create automated software developers, web researchers, and related back-office workers.

The concepts are rooted in good engineering principles. Every day professionals break down their work into small, manageable chunks and check their work when finished. Except now it’s possible to perform many simple tasks using language models. In my own experience with software development, using AI for basic tasks is faster and cheaper. Although it demands patience at first, as all new tools do.

Two of the most important startups in this space are Llama-Index (for database) and Langchain (for language model orchestration). Another project you should know is Ollama.

In my opinion, the most challenging task is orchestration. Language models by themselves are incapable of performing work to desired specifications because there is no way–yet–to replicate human will. That is, models are not guns: You can’t just point them and fire and expect to hit your target. You must “prompt” them, handle errors, and manage when things don’t perform as expected. It’s sort of like training a new employee. Professionals, your jobs are not in danger.

While imperfect, these systems show substantial speedup of tasks that humans perform slowly, such as web research and basic programming. In the new era of ai-assisted professional work, a professional will leave these tasks to an AI agent, which they will guide to a desired end state. That leaves more time for innovative, creative tasks.

At the current rate of progress, I expect most simple back-office work will be automated in a few years. As capital from these areas is freed, a portion will be returned to shareholders, while a larger portion can be reallocated to the all-important intellectual capital and mineral exploration needed to further one’s business.

The importance of preparing yourself and your organization cannot be overstated.

“‘Imagine we discovered a new continent with 100 BILLION people on it,
and they’re all willing to work for free!’ That’s what’s about to happen with #AI, so you’d better factor that into your plans.”

Related News

🤔 Opinion

Two topics have occupied my mind lately. The first is how to augment the talent shortage in mining with AI agents. And the second is macro economics.

It’s been repeatedly stated in recent trade publications that the mining workforce is both aging and shrinking without enough skilled replacements. How is an industry whose demand is expected to multiply many times over in the coming decade supposed to deliver without new talent?

One potential solution is targeted AI agents–systems which can perform professional tasks and be supervised by mining superstars. This is the area I’m building today. Lately I’ve been experimenting with automating complex earth-observation tasks as well as burdensome regulatory paperwork. (I’m looking for a few beta testers if you’re interested.) I think this area holds great promise for many industries, especially mining.

Next, the markets … they’re piquing my interest lately. I’m seeing a setup for a reversal in price trends from both large mining companies and commodities futures.

There has been a recent uptick in commodity prices, including lithium, cobalt, nickel, and copper. While the former 3 are still far down from their highs, I wonder if prices have reached their floor.

Overall, I don’t try to explain WHY the market has reacted the way it has, which I believe is impossible to do. But I do try to understand HOW it behaves. I’ve said before that the mineral markets have a volatility profile similar to that of bitcoin where it’s typical to see sharp rises and falls in a short timespan. This is good news for traders who seek volatility. I am glad I stayed out of the recent chop that has characterized the past year or so. Now I’m beginning to wonder if we’re approaching an inflection point in the economy and a good time to step in.

I am preparing for one of two scenarios:

  • Scenario 1: The mineral market confirms a reversal of its latest downward price trend, in which case I intend to play the markets in a big way over the coming years in tandem with the AI and electrification narratives.

  • Scenario 2: Lower lows are in store, compounded by the identification of alternative materials for use in battery electric vehicles. The recent upward trends are only fake-outs and buyers will be exhausted once sellers step in to rotate assets into other sectors of the economy.

Whether we reach higher highs or lows depends in the short term on one’s overall view of the strength of the world economy–and perhaps just as much on one’s view of any federal government’s ability to stimulate sustained growth.

It is clear from recent policy and legislative efforts in the US that the federal government seeks to foster mining, both domestically (recycling and refining) and through its partners (raw ore mining). Regulatory hurdles for domestic mining remain high, making US-harvested minerals significantly more expensive than overseas resources.

Recent efforts by federal agencies seem to favor the helicopter–money approach to its selected partners. And as a national security issue, US mining will likely see support through DoD funding as well.

Together, these suggest that traditional mineral funding sources will be competing with low-interest (forgivable?) government loans. While that’s good for miners, it may not be favorable to banks, which might seek greener pastures and thereby limit funding for new projects.

Of course, if Iran decides to attack Israel, all macro and technical analysis is useless. Long term, I am willing to bet technology improves, cars get cleaner and cheaper, and electric mobility advances across the globe.

Fingers crossed for the short term.

🛠️ Tools and Data

Tools and data to speed your work.

Python-based simulation environment designed for truck dispatching in mining operations to model and simulate various mining scenarios. Also read their Arxiv paper on simulating mining truck dispatching.

a framework for algorithmically optimizing LM prompts and weights, especially when Language Models are used one or more times within a pipeline.

Python framework showing how to use Google Earth Engine to track mining in the Amazon Rainforest.

Track positions and orbits of over 22,000 satellites orbiting the Earth in a 3D geospatial viewer powered by Cesium.

Thanks for reading! Want me to look into a particular topic? Email your suggestions and and I will dig.