Digital Exhaust #200

AI cannibalism, rising speech delay, and my father's imagined trip to the poor house

It’s another edition of Digital Exhaust, my weekly digest of interesting stuff from the web and beyond. A slightly shorter edition this week. Thanks for subscribing. And if you’re not one of us, subscribe just below.

Derelict insane asylums across America

Here’s a cool collection of abandoned asylums, state schools, hospitals, and TB sanatoriums across the U.S.. Some of the architecture is gorgeous.

Growing up, my father, Dick Vartabedian, used to say to my mother, “Dotty, you’re going to drive us straight to the poor house.” I thought it was just an expression. So I was thrilled to see that there actually is a place called “The Poor House.”

If anyone wants to go splits with me we could fix a couple of these up as Airbnbs.

New word: AI cannibalism

Users of GPT-4 model have been complaining about a dip in output quality.

One explanation: As large language models like ChatGPT and Google Bard scrape the public internet for data to be used when generating responses they are likely to learn from materials already produced by an AI. It’s the blind teaching the blind leading to a type of compounding bullshit known as AI cannibalism.

This runs the risk of creating a feedback loop, where AI models ‘learn’ from content that was itself AI-generated, resulting in a gradual decline in output coherence and quality. With numerous LLMs now available both to professionals and the wider public, the risk of AI cannibalism is becoming increasingly prevalent - especially since there’s yet to be any meaningful demonstration of how AI models might accurately differentiate between ‘real’ information and AI-generated content.

+ Related: I can’t disagree with Fast Company’s assessment that we are in ‘the age of algorithmic sameness.’

Big opportunity here for humans to do their weird, unique things.

Great stats and graphs on the healthcare labor market

From Peterson KFF is a comprehensive overview of healthcare labor stats drawn from the Bureau of Labor Statistics. The graphs here are wonderful. Tons of nuggets here.

While hospitals and physicians’ offices have returned to pre-pandemic employment levels, industries such as elder care and skilled nursing care facilities continue to see relatively low employment.

This Healthcare Dive article, Primary care providers say field is ‘crumbling’, echoes some of the practice wage stats in the KFF piece just above. Nearly 80% of respondents felt the current workforce is undersized to meet patient needs, and just 19% of clinicians report their practices are fully staffed.

Big picture on AI and the automation of work

Speaking of labor, I stumbled on this outstanding analysis on how ChatGPT and AI will change the workforce. This is from Benedict Evans, one of the smartest people in the room. Great read.

Rising speech delay

Interesting: Epic evaluated the rate of speech delay diagnosis by age two in 1,667,926 children. What they found is that children who have turned two years old since the start of the pandemic are more likely to have a speech delay diagnosis compared to those who turned two in earlier years.

Democratized neurosurgery

A Russian man claims to have implanted a chip into his brain in an attempt to control his dreams. He reportedly inserted the chip in his living room after watching neurosurgery YouTube videos and practicing on some sheep.

The finest medicine

“To be silent the whole day long, see no newspaper, hear no radio, listen to no gossip, be thoroughly and completely lazy, thoroughly and completely indifferent to the fate of the world is the finest medicine a man can give himself.” — Henry Miller

---

Thanks again for being a subscriber. Please pass this along if there’s someone who might be interested. Have a great week.