It's beer o'clock.
Founder JP's notes from the edges of innovation, at quitting time.
Jennifer Pierce, PhDFounder @ Singular XQ | I build the leaders we need for tomorrow. Transcendent Thinking.June 25, 2024
It's beer o'clock everyone but I thought I'd stop for a minute and show you what I've been working on lately.
Does synthetic data solve ethico-political problems with artificial intelligence?
In a word, no. My confreres who came up through human factors or ux research, you know the deal. Authentic primary data collection is costly and time consuming and no one wants to do it. Enter: synthetic data. I still haven't seen anyone actually publish useful, transparent tests of synthetic data that doesn't create bias, overfitting, and model instability. For the sake of argument, let's say that it could be done. Let's say, in theory this can be overcome through creating thoroughly vetted and robust data sets, this group of researchers says at best its a side-step from the issue and actually amplifies the ethical and political considerations around the idea of synthetic data sets. And call me crazy but doesn't vetting, reinforcing, scrubbing, and otherwise making synthetic data more robust sound just as costly as doing actual, human research? Work appears to be a fundamental need in our human systems that never appears to go away. It just gets displaced to others, usually those who are forced to forgo fair pay and human rights in other parts of the world.
Do LLMs present unique, unexamined threats to data sovereignty?
Kang Gu wrote a kick ass dissertation for Dartmouth examining what has appeared obvious to many of us: LLMs create unique opportunities for people to extract PII from embeddings. Focusing primarily on situations where the "fine-tuning" and "customization" being done on enterprise LLMs involve propriety data, Gu and his team find significant opportunities to recreate the original data sets. Gu explores methods of unlearning as a potential solution. In the meantime, 64% of businesses are significantly working LLMs into the business operations. Is there any wonder that we are under constant cybersecurity ransom attacks currently? LLMs are attack vectors, people. It's as if they were designed as a gift to hackers. Now PII is sorted by color, size, and shape for easy shopping experiences for those in the market for stolen data.
Who is going to see the Confucius Code?
I have never been to China but if I were to go suddenly this would be top of my itinerary. As a student of large scale multi-media performance art that mixes the physical and the digital and a former philosophy major with a taste for the philosophy of the East, the Confucious Code looks awe-inspiring.
Do autonomous robots in public parks freak people out?
Yes, Virginia. Yes, they do. The presence of robots in public spaces post Covid-19 has created such negative sentiment that it threatens the positive health effects such spaces create in the cities where they are funded and conserved. What's the answer? Human-centered design, of course.
How does AI hype work?
It's about the emotions, stupid. I keep wondering when people will realize that economics and business is driven as much by emotion as it is by numbers and dollars and cents. This performs an analysis of affective capitalism through AI hype. More on "affective capitalism" another time.
Happy Tuesday, y'all. And happy beer o'clock.
