• Drifft Newsletter
  • Posts
  • DRIFFT Newsletter - pipelines for geospatial ai workflows and who profits most from AI?

DRIFFT Newsletter - pipelines for geospatial ai workflows and who profits most from AI?

read in 5 min

Hey – David here. In this edition of DRIFFT you’ll get:

💡What’s New: Pipelines for geospatial workflows.

🤔 Opinion: Who profits most from AI?

🛠️ Tools: Ways to speed your work.

🎓 Tutorial: I used AI to beat a $500 mistake.

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

💡What’s new

There’s a lot of talk in AI today about foundation models. But, IMO, there’s too little discussion of foundation pipelines.

Pipelines are how you train and compose AI models. If you can manage this yourself–say for a mining company to monitor its own assets, or for an individual to prove an investment thesis–then you aren’t dependent on other expensive solutions.

…but…

Setting up these pipelines is difficult. Mastering the common elements of pipelining is often the mote that creates value. (To dig into that point, read Andrej Karpathy’s article on Software 2.0.)

That’s why I was really impressed to see several recent opensource projects focusing on geospatial AI workflows. I’m thinking specifically about Zen3Geo and Iquaflow.

Ironically, most foundation AI models these days are free. Meanwhile, most enterprises don’t know how to build complex pipelines from scratch. These kinds of projects are bringing geospatial AI workflows much closer to being consumable by a general audience.

I’m keeping my eyes open for more of these jewels.

🤔 Opinion

AI is called a transformational technology. So what will it transform?

Mostly existing businesses.

My experience building AI at BMW convinced me that it is not as earth shattering as most are claiming. Before you disagree, please let me explain.

In my last post, I said that AI is to the 21st century what the tractor was to the 20th. It does change things. It accelerates work.

Here’s who AI really helps.

AI helps incumbents more than newcomers. It helps existing businesses and individuals with existing cashflows by making them more profitable. It makes experts faster, more efficient.

I saw this at BMW. And I’m seeing it everywhere today. You’ll get the most out of AI if you already know how to program.

Will we see major enterprises emerge that focus solely on AI? Sure. But not as many new companies as the .com era gave us. (Someone correct me if there’s a study that suggests different.)

Sure, there are the NVIDIAs of the world that command a major portion of this new computing market. But Jensen has been building that business for around 20 years! AI takes time.

Should you be worried AI will devour your company with robots? The markets sure don’t seem to think so. Or at least they haven’t priced that eventuality into their stocks at this point.

So neither will I.

🛠️ Tools and Data

Opensource to speed your daily geo tasks

  1. Zapier - a staple for automating workflows between different apps, now supporting GPT integration. https://zapier.com/

  2. Geemap - Google’s earth engine in the Jupyter notebook, and supports Meta’s Segment Anything model. https://geemap.org/

  3. Iquaflow - a set of tools and models from Satellogic for assessing satellite image quality. https://github.com/satellogic/iquaflow 

🎓 See What’s Possible

One thing every day for 365 days makes you unstoppable.

Here’s how I recenlty used GPT to save $500 on medical bills because of a misdiagnosis.

What went down?

A physician misdiagnosed a condition and charged me for the service. “That’ll be $500+” for the tests. But the problem wasn’t fixed, so I wasn’t convinced I should pay. (Yes, I expect results from ALL my vendors).

Next, I ordered my test results and found abnormalities in the data. I’m no medical expert, soooo…

I hand annotated the images and sent them back. And this time the doc was forced to look. They agreed there was a problem. Big whoop. I was still stuck with this janky bill.

That’s when I turned to ChatGPT.

Per ChatGPT’s recommendation, all personal data was scrubbed from the inputs. But it works just as well with fake data as with real.

The letter it produced was actually much longer and more eloquent than the screenshot shows. But you get the picture. In 2 weeks I had my answer. “We’re sorry. We’ve removed the bill.”

I imagine you can do something similar when vendors try charging you or your company for inadequate results. Good writing takes a lot of time. And hiring an attorney is mostly an unattractive prospect.

I have trialed https://donotpay.com/, which is a self-described AI lawyer. Though I have not field tested and will not endorse at this time. But based on my results with (free) ChatGPT, I’d say I’m definitely looking more closely at these solutions.

Free capital is a major asset in a down economy. Any time you can save $$ is a win in my book.

Go get those savings!

My favorite map graphic from the week.

From the Journal of Maps, 2023-05-01. Geologic structures for an entire continent classified using Google’s Earth Engine and satellite imagery.

Geologic structures for an entire continent classified using Google’s Earth Engine.

Share DRIFFT

Know someone looking to level-up their AI and opensource game in mining? Just copy and paste this link: https://drifft.beehiiv.com 

Thanks for reading! I am here for you. Want me to look into a particular topic? Email your suggestions and feedback.