Opensource AI and which areas you should take seriously

💡What’s New: Opensource AI – an intro to which areas matter

🤔 Opinion: Will Ford’s EV pullback clobber mineral markets?

🛠️ Tools & Data: Fantastic opensource tools to propel your work.

Want to feature your service or product in DRIFFT? Grab an ad spot here.

💡What’s new

This week I bring you an overview of opensource AI and the areas seeing the most development. This will start a deep-dive series on utilizing AI to get things done.

Modern AI is unnecessarily complicated. It is nothing more than a collection of statistical models for pattern recognition. “Deep learning” is the most popular sub-field today that is nothing more than applied calculus with computers. It is just a set of logical tools. (The very best technical synopsis I have ever read is in chapter 2 of Machine Learning for Risk Calculations.)

The areas of AI receiving most development dollars today are:

• Language - The AI models are called “large language models”. GPT is the most popular. They are really good at predicting the next element (word or paragraph) in a sequence, and useful for summarizing content. The most prolific opensource version is Llama by Meta. They’re “large” in the sense that a model is comprised of many billions of parameters, giving it a large computer memory footprint.

• Vision - These models allow computers to process visual information. They are also large, like LLMs, but they’re used to predict pixels in an image rather than words in a sentence. They’re especially good for tasks like identifying and locating objects in an image.

• Multimodal - These models combine multiple data “modalities” (e.g., language and vision) so you can work between types of data. For example, you can create images by describing them or even caption images and video. The latest models are being used to make robots perform actions simply by describing what you want them to do. Multimodal models are the engines powering most AI startups today.

• Expertise - These models learn from the behaviors of “experts” in a process called “reinforcement”. That just means that the output of a model (described above) is combined with the judgement of a human expert in order to adjust the system to perform at expert levels. For example, by identifying skin cancers or identifying mineral elements.

Several applications for mining are seeing significant development and will impact your future.

• Earth observation (satellite, drone, rover) - images can be used to detect minerals, monitor operations, and perform long-term environmental studies. Refer to some of my past newsletters for examples. Historically these capabilities were only available to governments but are now low-cost and accessible to anyone with internet. The ability to monitor any position on the planet in near real-time will significantly impact ALL industries. Business results will come from task speed ups, fewer accidents, and reduced exploration risk.

• Task Specialization - You can “train” an opensource model to become an expert on a specific routine task in order to speed it up and/or reduce the need for costly experts and custom software. For example, you can train Llama to follow the news and extract meaningful insights about competitors.

• Task Orchestration – It is possible to create AI work crews that do work for you. See below a recent tutorial I made on this topic. IMO, this application is the most critical for dealing with the short-term challenge of finding more mining professionals. With AI agents it’s possible to realize significant speedups in performing common tasks while reducing the cost of those tasks by an order of magnitude.

– – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – – –

Related News 

• Use ChatGPT to learn geospatial programming

• A Stanford AI postdoc takes a job in mining.

• The many use cases for AI in the mining industry.

• NASA satellite uses airborne mineral particles to detect where they came from.

• AI costs are hindering widespread adoption.

• Geospatial imagery analysis is growing fast – ~$73B by 2030!

• “The era of critical minerals is an era of reinvention.”

• The S&P talks use cases of blockchain for commodities.

🤔 Opinion

Ford and Chevrolet have recently flip-flopped on their commitment to electric vehicles. Ford has “pushed pause” on billions in investment. Combined with continued high interest rates, poor sales data, and broader economic cooling, is this a signal of worse to come in the battery metals markets?

My opinion–we’re in a protracted lull (not a bull, not a crash, but something in between).

Today there’s a 12-month backlog of EVs sitting on dealer lots at a time when most consumers are struggling to buy their groceries. GDP is up, but primarily due to inflation as consumers must spend more. I have spoken on Twitter about the fact that Chinese manufacturers are outpacing companies like Tesla, BMW, and Mercedes. When a consumer can get all the luxury and features they want in a car for half the cost, the only ways to prevent a takeover are tariff / export wars, which have already proven to be inadequate measures. If / when ICE engines are outlawed–as they are set to be in many US and EU cities–then the currently subsidized price for BEVs will soar, making these vehicles even less appealing.

Market participants for major mining stocks seem to be consolidating positions–bears and bulls betting on the future of the broader economy. I keep asking myself whether we can expect to see the same downward slide that the raw commodity markets have recently suffered. This would make sense from past data. In my mind, all the buyers are exhausted. While news is saying, “buy buy buy”, markets seem to be showing that there’s more surprises to come.

So I remain short-term bearish. Without a market-wide capitulation, a drawdown of ~60%± from the highs, I am waiting to see more data.

It’s disappointing to see automakers and consumers on tough times. But the market rarely gives us exactly what we want. This lull can provide space for other organizations (especially well-financed startups) to take up the unused resources and innovate, which is also needed.

When the situation balances, I believe it will be a net positive for our economy and the mining industry. I am still long-term bullish.

🛠️ Tools and Data

Opensource solutions that will inspire you to create new value.

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

David