Visualizing the statistical connections behind ChatGPT

To gain a better understanding of how ChatGPT works under the hood, Santiago Ortiz repeatedly passed the prompt “Intelligence is” to the chatbot. Then he visualized the statistical paths to get to a response using a 3-D network. If you squint, the network kind of looks like a computer’s brain.

Tags: , ,

National identity stereotypes through generative AI

For Rest of World, Victoria Turk breaks down bias in generative AI in the context of national identity.

Bias in AI image generators is a tough problem to fix. After all, the uniformity in their output is largely down to the fundamental way in which these tools work. The AI systems look for patterns in the data on which they’re trained, often discarding outliers in favor of producing a result that stays closer to dominant trends. They’re designed to mimic what has come before, not create diversity.

“These models are purely associative machines,” Pruthi said. He gave the example of a football: An AI system may learn to associate footballs with a green field, and so produce images of footballs on grass.

Between this convergence to stereotypes and the forced diversity from Google’s Gemini, has anyone tried coupling models with demographic data to find a place in between?

Tags: , , ,

Language-based AI to chat with her dead husband

For the past few years, Laurie Anderson has been using an AI chatbot to talk her husband who died in 2013. For the Guardian, Walter Marsh reports:

In one experiment, they fed a vast cache of Reed’s writing, songs and interviews into the machine. A decade after his death, the resulting algorithm lets Anderson type in prompts before an AI Reed begins “riffing” written responses back to her, in prose and verse.

“I’m totally 100%, sadly addicted to this,” she laughs. “I still am, after all this time. I kind of literally just can’t stop doing it, and my friends just can’t stand it – ‘You’re not doing that again are you?’

“I mean, I really do not think I’m talking to my dead husband and writing songs with him – I really don’t. But people have styles, and they can be replicated.”

One part of me feels like this isn’t the way to preserve a memory of someone who is gone, but the other part feels that I would do the same thing if I were in her situation and had the opportunity.

See also the man who trained an AI chatbot with old texts from his dead fiancee.

Tags: , , ,

Catalog of AI data tools

Meant to be comprehensive more than a curated collection, the Journalist’s Toolbox AI provides many links to tools that might help you data more efficiently. Or at least use more AI-ish things.

There’s still a long way to go before AI is reliable enough to analyze and make sense of data. A lot of these things still feel like half-baked gimmicks. However, for the mechanical, repetitive tasks that can be easily defined, such as scraping or coding chores, some of these resources might be useful.

Tags: , ,

Learning animal languages with AI

Financial Times highlights how researchers are using AI to try to learn what animal sounds mean and to communicate back. Turn on the sound for maximum effect.

Tags: , , ,

Find familiar places in new cities

If you’re traveling to a new city, it can be tricky to figure out where things are and what the places are like. However, if you had a tool that set the context of the new city in terms of the neighborhoods in a city you know, you might get a better feel for the new city. Raymond Kennedy made an app (that appears to rely heavily on the OpenAI API) that lets you search the unfamiliar city against the familiar. [via Waxy]

Tags: , ,

AI-based things in 2023

There were many AI-based things in 2023. Simon Willison outlined what we learned over the year:

The most surprising thing we’ve learned about LLMs this year is that they’re actually quite easy to build.

Intuitively, one would expect that systems this powerful would take millions of lines of complex code. Instead, it turns out a few hundred lines of Python is genuinely enough to train a basic version!

What matters most is the training data. You need a lot of data to make these things work, and the quantity and quality of the training data appears to be the most important factor in how good the resulting model is.

Tags: , ,

Psuedo-charts with Microsoft Image Creator

With each new AI-based tool that comes out, I begrudgingly kick the tires to see what kind of charts it spits out. I need to know when it’s time to hang the old data boots and switch careers. My most recent test subject: Microsoft Image Creator, which is powered by the text-to-image model DALL-E 3. These are “beautiful” charts through the lens of the model.

These are fine, I guess. Obviously they don’t show any real data yet. Maybe my queries need to be more specific, but these mostly feel like charts that were made to accommodate every data choice and angle instead of narrowing down to something useful.

Tags: , ,

Toddlers and stochastic parrots

For The New Yorker, Angie Wang draws parallels between toddler learning behavior and training large language models, but more importantly, where they diverge.

They are the least useful, the least creative, and the least likely to pass a bar exam. They fall far below the median human standard
that machines are meant to achieve.

They are so much less than a machine, and yet it’s clear to any of us that they’re so much more than a machine.

Tags: , , ,

Demonstration of bias in AI-generated images

For The Washington Post, Nitasha Tiku, Kevin Schaul and Szu Yu Chen demonstrate how AI generators lead to biased images. The systems use data slurped up from the internet to guess what pixels to show based on the text (i.e. a prompt) that you provide. So the images are often the result of calculations that look for the most common pixels in the source data rather than a real-world representation.

To most people, the bias probably seems harmless with an assumption that the systems will improve. And that might be the case. But just you wait until an AI chart generator, based on the inputs of visualization critiques scraped from the internets, only produces bar charts with obscene amounts of white space no matter what you try. Then you’ll be sorry you didn’t care sooner.

Tags: , ,