Making AI Productive pt 4 - Why Generative Workflows

💡What’s New: Models in reasoning pipelines - workflows for professionals.

🤔 Opinion: The mineral investment narrative is changing.

🛠️ Tools & Data: Tools and data to speed your AI transformation.

💡What’s new

In the last post, we reviewed AI for images. These image models, along with large language models, form the bulk of today’s AI. But the focus is beginning to shift to generative AI.

Because most professional work consists of reasoning over text and images, and since we already have excellent models for those data types, the next logical step in AI is to automate them. Today we look more closely at generative AI as assistants and the changes coming to professional work.

In part 2, I briefly touched on AI chains and agents. These form the basis of generative assistants, so-called because they generate responses to our inputs as, say, text or images or even movies. But to understand why they’re a big deal for professionals, we need to get into the weeds.

In its last major GPT release, OpenAI introduced the concept of “function calling,” which allows developers to use ChatGPT to generate properly formatted inputs to their code when the AI concludes running a piece of code is the right action. (Note: This concept was not unique to OpenAI, however, and many opensource models have matched and improved on this capability.)

I cannot overstate how significant function calling is–not only to the software model, but also to generative assistants.

Throughout the 1990s and 2000s, a lot of valuable software platforms worked on a simple template. A company’s product consisted of a set of packaged capabilities and sold under a brand name. So, for example, Apple transformed Linux into its Mac OS (Unix) and made gagillions. Within this model, the professional worked step-by-step through the software to accomplish their tasks.

Generative assistants change this model. OpenAI demonstrated that AI can code common programming tasks and even functionality it has not seen. And with function calling, it showed AI can even run the code and subsequently interpret results.

So we have:

1. Reasoning over text and images,

2. Ability to program,

3. Ability to integrate with other software tools,

4. Ability to plan and complete tasks involving the last 3.

This is the generative workflow. And it is infinitely customizable to any work.

In my own experiments, my AI chains and agents running on my laptop can integrate with tools and existing services like a Python REPL to run code, Github to manage it, Google Search to get information, and HuggingFace to implement other models as tools.

With generative workflows, the seed has been planted for a new era. Building software will no longer be the primary focus of tech companies as that task can be automated. Instead, they will package generative workflows that provide the ability to get real work done fast and with greater quality at a much lower price than was ever possible before. It is beginning with software and will transfer into increasingly complex tasks.

Fellow professionals, the work model has changed. Our jobs are becoming more interesting and more complex. Now the professional can be in total control with no more programmer intermediaries. Instead of manually doing work with packaged functionality, professionals can be building agents and managing their outputs.

A friend recently challenged me on this idea and asked, “Do you really think this model can replace something as simple as Amazon product search?” In other words, isn’t it faster to type into a text box and get the results like we’re already used to? I don’t know. But I know both Google and Amazon recently launched AI assistants to try.

What does this mean for professionals in industries like mining, earth observation, and geosciences? They will soon discover what the rest of us in AI have known. It’s possible to automate complex professional tasks. And it is even possible to side-step paying for software by having an AI build it for them.

Generative workflows open a wide door for creative exploration of complex problem spaces that before were either too costly or which we had no time for. I envision a future of vastly more mineral data, customized software, automated satellite observation analytics, and automated mine monitoring–all at much lower costs.

It will take time. But not as much time as the first software revolution.

If you’re a professional or an executive wondering how you should position for these changes, I suggest learning how to utilize opensource AI. Generative workflows will offer every professional more productive careers. Also, the ability to easily create and experiment means that we will all face more competition. And it will be imperative in the short-term to position ourselves for these changes by adopting technologies that let us focus on creating rather than managing.

For my part, I’m building generative AI workflows for the mining industry. Some of them I’ll opensource, so stay tuned.

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

Related news

• alwaysAI enters the mining industry selling computer vision applications.

• Are commodity company softwares more valuable than commodity companies?

• A great overview of how mineral companies are currently using AI.

• Stanford and FleetSpace team on micro-satellites that can map the subsurface in days

• The US is now partnered with Kazakhstan on mineral exploration with AI.

• Caveats on using the Inflation Reduction Act to finance mining.

• Australia’s Northern Territory is pushing $3M (AUD) to fund mineral exploration.

🤔 Opinion

Anyone who has looked at BEV markets with both eyes recently may have come to a conclusion similar to mine–there are dark clouds on the horizon. Sales in Germany and the US are in decline. The economics for most US consumers considering major purchases aren’t improving either.

Will this spell doom for minerals markets and startups?

Just a few months ago, the euphoria over battery metals was unbelievable. The sentiment across the internet was mega bullish. Even while commodity futures prices and multinational stock shares continue a recent and sharp bear trend, it’s easy to find Twitter analysts calling for quick recovery and all-time highs. (I don’t care whether they’re right or wrong.)

Today’s narrative on the catalysts for mineral valuations is shifting. Battery minerals have become critical minerals. BEV shortages have become Chinese rare-earth export restrictions.

I agree that long-term there appears to be an opportunity. While the US will likely not become completely mineral independent, it does appear it wants to be less-dependent. (All the more important if the BRIC nations manage to shake off of the dollar collar.) Lately the US has been knocking on the doors of Australia, Chile, Kazakhstan, and Indonesia. Money is being spent.

But short-term, the BEV demand catalyst is uncertain. And government support is at the whim of changing administrations.

The current situation reminds us why it’s important to stay agnostic to any narrative. Great investors don’t try to predict. They position. They play the odds that are in their favor. They use opportunities the market gives instead of holding to any dogma–the market must do this or that. The market is not predictable.

This means that whether I was running a startup or a hedge fund investing in minerals, I would use everything today to my advantage. I would use BEV or critical minerals. I would use government subsidies and laissez-faire pushback. I would use climate activism and climate deniers. It’s all opportunity.

Stay sharp. The only thing we can guarantee in our markets is our own actions.

🛠️ Tools and Data

Opensource tools and datasets to speed your ai transformation

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

David4