Lightning Rod Physics

"Striking Discovery: Unique Image Reveals How Lightning Rods Work". (Via H.R.)

 

Entangled Photon Update

"Entangled Photons Produced Entirely On-Chip"

Eyespot Evolution

"'Mona Lisa effect' applies to animals, too"

Browser Password Managers

"Warning: Don't Let Google Manage Your Passwords"

GPS Tractor Fail

"Farmers 'crippled' by satellite failure as GPS-guided tractors grind to a halt"

Square-Wheeled Bicycle

"Square-Wheeled Bicycle Looks Like An Optical Illusion Come To Life"

Wolverine Software

"Developer creates 'regenerative' AI program that fixes bugs on the fly"

Robotic Dog On Balance Beam

"CMU taught a robot dog to walk a balance beam". (Via H.R.)

Google And AI

"Google's looming AI integration into Search should scare the hell out of Microsoft"

Light Posting Notice

Admin note: Posting may be lighter than usual this week due to external obligations.

Apple Lisa Revisited

"This video shows what it's like using an Apple Lisa from 1983 in 2023"

Checking ChatPDF Accuracy

Scientist tests ChatPDF to see how well it "understands" the content:

For this test I'm going to take the five most recent papers on this blog and use the specialised ChatPDF. Let's see if this does better than the ChatGPT app and how it compares to doing to hard work of actually reading the paper...

Conclusions

I'm afraid this one can't be said to be anything beyond the usual "impressive tech demo" stage. It is categorically not ready for actual use and anyone paying the subscription fee is at this stage a complete fool. 

To give credit where credit is due, it does often produce remarkably good summaries that are more accessible than reading the abstracts. It can extract complex variables, even ones which are stated directly in the text. It seems to do better when you ask it very specific questions, but it's capable of handing complicated technical descriptions and distilling them down to their most relevant points in even more generalised ways.

The problem is that accuracy and usefulness do not scale linearly with each other. If it produces accurate statements 70, 80, even 90% of the time, it's only useful 0% of the time. Why ? Because that failure rate is such that its claims always have to be checked, and one would be better off just reading the paper. You have no idea if it's just making stuff up or missing a vital point. Worse, it's dangerously coherent. If you're not already an expert in the field, it produces statements which sound fully convincing but are in fact just plain wrong. I'm glad it references the parts of the text it's getting its information from, but it frequently just invents entire quotes, and that's unacceptable.

 

AI Generative Agents

"Generative Agents: Interactive Simulacra of Human Behavior"

In this paper, we introduce generative agents--computational software agents that simulate believable human behavior. Generative agents wake up, cook breakfast, and head to work; artists paint, while authors write; they form opinions, notice each other, and initiate conversations; they remember and reflect on days past as they plan the next day. To enable generative agents, we describe an architecture that extends a large language model to store a complete record of the agent's experiences using natural language, synthesize those memories over time into higher-level reflections, and retrieve them dynamically to plan behavior. We instantiate generative agents to populate an interactive sandbox environment inspired by The Sims, where end users can interact with a small town of twenty five agents using natural language...

Related Ars Technica article, "Surprising things happen when you put 25 AI agents together in an RPG town".

Mario Theme Honored

"America cements Super Mario Bros. theme as part of nation's history"

[T]he Super Mario Bros. theme has been rightfully recognized as a key contribution to U.S. history. It’s one of the 25 songs that will be added to the National Recording Registry this year, Librarian of Congress Carla Hayden announced on Wednesday, and the first video game song to ever be added. 

The Super Mario Bros. theme, officially titled “Ground Theme,” was composed by Koji Kondo, the storied Japanese composer behind many of Nintendo’s hits. The theme, released in 1985, has become ubiquitous — it’s appeared in numerous subsequent Mario games, was a theme in the recent Super Mario Bros. Movie, and has been remixed countless times across YouTube, TikTok, and elsewhere.

AI-Assisted Fake Kidnapping

"'I've got your daughter': Mom warns of terrifying AI voice cloning scam that faked kidnapping" 

In this case, the mother was able to separately confirm her daughter was ok before paying any ransom.  

The development of this technology adds more immediacy for people to find ways to authenticate that any call supposedly from a loved one or friend is truly from that person, especially in an apparent crisis. (Code word, shared private memory, etc.?)

Beating Roulette

Applied physics: "The Gambler Who Beat Roulette"

More Edge Features

"Microsoft's rolling out Edge's AI image generator to everyone"

Popular Songs 1980 Onward

"The Most Popular Song From Each Month Since January 1980"

Pine Cone Art

"This mesmerizing pine-cone installation is made from 95 reused car hoods". (Via H.R.)

 

Rainbow Shape

"Rainbows are actually full circles. A physicist explains"

AI Fashion Models

"Levi's to use AI models to 'increase diversity and sustainability'".

If I were more cynical, I'd say this was a way to pro-"diversity" public image without having to pay for actual human workers who meet the desired demographic profile. 

 

AI "Hallucinations"

"Why ChatGPT and Bing Chat are so good at making things up"

Monkeys And Magic

"Sleight-of-hand magic trick only fools monkeys with opposable thumbs"

Analog Computers, 2.0

"The Unbelievable Zombie Comeback of Analog Computing"

Hiccups And Cancer

"Hiccups have a curious connection to cancer"

Stop Motion Animation

"How Stop-Motion Movies Are Animated at Aardman"

AI And Copyright Update

"Stable Diffusion copyright lawsuits could be a legal earthquake for AI"

Silent Mode Haircuts

"A new barbershop in San Francisco offers silent haircuts so that you no longer have to talk to your barber as they line you up."

At Beyond the Pale barbershop in the Mission District, there’s no need to say a word with the shop’s “silent mode” service, which was designed especially for shy techies and stoners, according to shop owner Anthony Larrasquitu. 

“Ultimately, it’s about the freedom here. You get to pick the experience you want,” Larrasquitu said.

21st Century Relationships

"China's Newest Dating Craze: Real-Life Meetups With Virtual Boyfriends"

Too Much Tipping

"Tipping in the United States has gotten out of control, experts say. Here's why"

Another reason people are tipping more is because of newer and cooler-looking technologies — kiosks and tablets with three large tipping suggestions that pop up on the screen in front of you. Business owners typically pick those options, and they can also disable the feature if they want to.

To that point, 22% of respondents said when they’re presented with various suggested tip amounts, they feel pressured to tip more than they normally would, according to Creditcards.com.

 

Hsieh Forbes Column: What Does It Mean To Say A Medical Test Is Sensitive Or Specific?

My latest Forbes column is now out: "What Does It Mean To Say A Medical Test Is 'Sensitive' Or Specific'?"

Some of the key concepts go back to World War II, and the early days of radar operator detection.

For a more in-depth discussion of these issues, as well as the closely related concepts of “positive predictive value” and “negative predictive value,” I highly recommend this excellent graphic novel (PDF version) by Dr. Stefan Tigges, Professor of Radiology at Emory University School of Medicine in the Journal of the American College of Radiology.

(Bonus: Dr. Tigges provides one of the clearest explanation I’ve ever read of so-called "Receiver Operating Characteristic" curves -- another concept which harkens back to the old days of radar operators in the 1940s.)

Better Financial Math Tools

"Mathematician uncovers methods to shrink sampling errors in large-dimensional data sets":

Kercheval's research provides a way for the analyst to better estimate the future risk of proposed stock portfolios by reducing statistical uncertainties, and this new method is most useful to financial portfolio managers who often run into challenges when determining financial outcomes for their clients when the number of assets held in a single portfolio exceeds the manager's possible observations.