It's beer o'clock.

What does data democratization look like in practice?

Do you remember Society 5.0? This group of researchers look at it as a roadmap from getting us over perilous data into true democratized data structures. It's a high level look but is good to reminder that between centralized data silos and a democratized data-powered future, lies even more perilous territory as the data moves out of silos into too few hands, without enough guardrails or security protections around it. Blockchain does hold some promise but people must get over a naive understanding of its immutability if we are to begin to leverage this technology for greater data sovereignty.

Little patent on the prairie...did you know that farmers used to have patent-breaking parties?

Patent and IP law are becoming more relevant to today's digitally transforming world and a lot is being decided this year in courts that will have sweeping implications. As a result I've been researching patent and intellectual property and history. We tend to think our world is so new and complex but usually there are historic analogs that inform our present day struggles while also proving that very little is new under the sun.

The first major attack on patent law was led by the Prairie Populists in the 19th century, as I learned from this well-written book chapter by Steven Wilf. Then, the patent system unfairly benefitted the wealthy and powerful as technologies farmers had used previously became protected under patent law.

A specific example was a Wire Binder which was a simple technology that had tremendous advantages to binding large amounts of grain for resale, quickly. Their efforts to stop the McCormick company from stifling growth and innovation amongst farmers who wanted to fashion their own versions of the machine (for far less money) was voiced so the courts would be less solicitous of patent holders and more supportive of small farms.

How much government money goes to the tech sector? I'll tell you Monday.

Fast-forward to today, we still have a patent system that benefits the powerful and the wealthy against the small. However, now patent infringement is a David and Goliath problem. Today, the small business and individual patent holder is the target of strategic infringement making the effort to innovate in small businesses and in open source modes more risky for the little guy.

I've got a book on my desk I might try to read this Fourth of July (one of my favorite boozy fudge around and find out holidays to read and play games) by Mariana Mazzucato called The Entrepreneurial State. I can't say what I think about her argument, but the idea that the government should stay out of innovation and let the robust, more daring private sector handle it is probably one of the most annoying sentiments I run into in emerging tech. In truth, I won't tell you how much of your tax dollars go to the innovation efforts conducted by the titans of tech. It might ruin your Independence Day buzz.

How fuzzy is the math in LLMs? [Shout: how fuzzy is it?]

The math and LLMs are so fuzzy...I'm pretty sure they use their fingers to count and most of them wear mittens.

(Did you know Fozzy bear is my Discord avatar? Yup. That's right. Goals.)

In a hashtag I'd like to call #theoceaniswetandothernews here is yet another article that shows how the fuzzy math of LLMs precludes any claim of emergent or possible reasoning in models so built:

LLMs fail significantly on problems with ambiguous and mixed scope, revealing critical limitations and overfitting issues (e.g. accuracy on GSM8K drops by at least 50\%).

They demonstrate a tactical training set that may help in the future but there are multiple other problems. In matching existing data in the data set LLMs will never be able to handle ambiguous or mixed scope problems with any reliability. It's a built in limitation to the model, even with the most advanced DNNs. Come on. We new this about ANNs first and we thought that multiple layers of noded networks would help It didn't. DNNs are still plagued by non hierarchical data sets and have increased energy requirements, higher risk of overfitting, and have difficulty interpreting learned representations within in decision- making.

Where can we find human rights positive hardware? The search continues.

The EU is leading the way in business ethics across the board and human rights issues in manufacturing is no exception. The research and scrutiny is already bringing changes and the ICT sector is making sweeping changes as well as using blockchain to promote transparency. Could a human rights positive hardware choice emerge? Time will tell. What an excellent start up idea. It's far easier to start with human rights in mind, than to have to disentrench human rights negative practices from legacy organizations who have relied on looking the other way o human rights abuses in order to hit their profitability targets. We have a top four but we would love to partner with a smaller, newer, more contemporary hardware producer. We are looking at Purism, SPC , System76 Novena and Fairphone but we are impressed with Dell Technologies and 华硕 for their recent efforts to improve their human rights' records.

I'm this much closer to no shoes and a drink in hand with a pile of books and some games and puzzles to play with the fam.

Happy Tuesday!

"It's beer o'clock" is not the official newsletter for Singular XQ but my own notes about what I've been researching from week to week. It is published on our ghost website along with our official newsletter. You can subscribe to all of it at singular-xq.ghost.io

Thanks to my subscribers and followers here. Seeing what you read and what you click shapes my research agenda more than you know.