Articles

How to use knowledge graphs with retrieval augmented generation (RAG) to organize large content repositories.

The Automation Paradox is that the better our machines get, the more we struggle when they fail. When heading out for a hike in the woods, it's tempting to skip the map and compass and rely on our phones and apps for navigation. Yet when we encounter no signal or lose power, we can find ourselves in a sticky situation. Or perhaps, like me, you've come to rely on popping your destination into the satnav or Google Maps whenever you get in the car and have nearly forgotten the ability to navigate without it. These situations illustrate the paradox of automation, where the more sophisticated and automated our machines and technologies become, the more bewildered we find ourselves when they inevitably fail. Or: the smarter the machines get, the dumber we might get. In his book Messy, Tim Harford suggests three strands to the paradox as our machines get more sophisticated: Automation covers up our mistakes, hiding our incompetence, meaning we may not learn to correct ourselves—consider autocorrect cleaning up our typos as we go. When we rely on automation, we get less practice for our skills, so even highly skilled individuals may find their expertise diminishing—perhaps you've found yourself using your phone calculator for a trivial calculation. When the easy scenarios are taken care of, failures may occur in complex or unpredictable ways that we may find especially difficult to recover from—like a subtle but persistent failure in the steering of a passenger plane, recovering from a skid on an icy road, or when you're deeply lost in the wilderness. More sophisticated technology can even make it useless or more dangerous when it fails. Older cars used to be reparable forever. Now, if your vehicle fails, it's likely to need plugging in at the dealership to figure out what's up. How many electronic devices are thrown away because somewhere inside, some tiny loose connection or component makes the whole thing worthless? Were pilots and flight crews better prepared and able to improvise before the autopilot became ubiquitous? I often wish our devices would fail more like an escalator or an electric toothbrush. If an escalator fails, you can still walk up it. If your electric toothbrush dies, you can still use it to brush your teeth. These could be called Technology-Enhanced Products, perhaps. But when most of our devices die, they're often rendered worthless. Automation and sophisticated machines help me so much. I did use Grammarly to help check this post. I use Google Maps nearly every time I put in my destination to home, and I often use a calculator to check my maths. But I do pay attention to grammar corrections, bring a paper map when I can, and keep trying to do the maths in my head. But as Tim Harford explains it, we still face "the paradox of automation: the better the machines get, the more bewildered we are when the machines fail." Also see: The bus factor Normalisation of deviance Chaos monkey Know your tech Jevon's paradox The law of unintended consequences The Dunning-Kruger effect (send me proof of purchase of Big Ideas Little Pictures, and I'll send the sketch to you) More paradoxes: The coastline paradox The transparency paradox The Abilene paradox The paradox of choice The liar paradox Tolkein-style landscape inspired by the excellent Lord of Maps.

And your free copy of 'A Glossary for the Appreciation of Life: Part One (A-K)'

Google developers for the open source language Go has started using Project Oscar to build AI agents to manage its issue tracking.

AI interview services say they’re eliminating bias — but not everyone agrees.

Episode 132

OpenAI's work seeks to give people a framework to train models to better explain how they arrived at particular answers.

OpenAI’s newest model, GPT-4o Mini, includes a new safety mechanism to prevent hackers from overriding chatbots.

The White House’s voluntary AI commitments have brought better red-teaming practices and watermarks, but no meaningful transparency or accountability.

Explore how technical and social structures co-evolved over five centuries in this large-scale research visualization.

We cannot let this fever run its course

When I first used Dall-E in January 2021, I input some text, I think it was a resin chair inspired by a cactus (I had Gaetano Pesce on my mind). It gave me an output that genuinely rocked me. I was testing this mental model that many people have for image generation that it’s basically just collaging existing images together and I realised fast that this was a very bad mental model. The first output blew my mind because it was so unexpected; It created a lot of detail within the image that I didn’t imagine when I wrote the prompt, which I could never find with a search engine. It had interpreted my prompt in a what I perceived as a

The goal is fungibility—to assert equivalence between activities by people or environments so that emissions created over here can be traded and (theoretically) compensated for by actions removing or reducing carbon over there. The means is, of course, commodification. Offsets privatize planetary metabolism.

New paper counts "excess words" that started appearing more often in the post-LLM era.

Welcome to the new post-genomic biology: a transformative era in need of fresh metaphors to understand how life works

Diane Coyle explains why, instead of simplifying daily life, technological innovation is making it more burdensome.

Online surveys have some key flaws.

Sam Altman and Arianna Huffington told me that they believe generative AI can help millions of suffering people. I’m not so sure.

Have you noticed artificial intelligence is looking a lot more sparkly lately? But first…

Amazon Employees for Climate Justice is calling out the tech giant for what it says are misleading claims about its use of renewable energy.

Carbon Brief provides an updated analysis of when the world will likely exceed the Paris 1.5C limit

In India's scorching summer heat, the ancient practice of chilling water in terracotta pots is inspiring new trends – from cooling towers to screens for buildings.

A new generation of forest defenders are fighting for their home in the face of extractive oil drilling and gold mining.

AI development has displayed signs of unchecked racism and environmental harm. Incorporating traditional knowledge might make it more just.

Nearly every iceberg you see in a picture or diagram is probably floating the wrong way. This was what I learned (from Megan Thompson-Munson) after sketching Biz Stone's brilliant saying about the myth of overnight success. With some approximation, the density of ice is around 900 kg/m3, and seawater is around 1,000 kg/m3. Therefore, the fraction of an iceberg that's submerged is around ~900/1000 or 0.9. So, about 90% of an iceberg is below the surface and 10% above, which is partly why they can be so dangerous. While most iceberg pictures get this part more or less correct, most of these icebergs will be floating vertically. In reality, a tall, thin iceberg will likely topple, so most icebergs end up floating on their side, not their tips, even though we rarely draw them this way. I remember learning about a fascinating experiment with children of different ages estimating which glass holds more water: a tall, slender one filled high or a wider glass filled to a lower level. Younger children almost always chose the glass with the higher water level as the most water, even when it was significantly less than the shorter and wider glass. I wonder if it's part of why we draw icebergs vertically, at least when we're using them as a metaphor. It's easier to grasp quantities vertically, and we commonly underestimate volume spread over a wider area. We may still want to draw our icebergs tall and deep to make our point, but now, at least, we can do so with the knowledge that they're not like the real ones. If you want to see it yourself, Joshua Tauberer made a brilliant draw-an-iceberg-and-see-how-it-will-float game. I recommend you give it a go so it sinks in forever (sorry). Also see: Overnight success Why ice doesn't sink Ice-cream, gelato, sorbet Know your poles: penguins or polar bears, frozen ice or land

Researchers from MIT and elsewhere developed an easy-to-use tool that enables someone to perform complicated statistical analyses on tabular data using just a few keystrokes. Their method combines probabilistic AI models with the programming language SQL to provide faster and more accurate results than other methods.

This post explores one way developers can go beyond page load and start estimating web page carbon emissions that include user interactions on the page.

I am a front-end developer who is FED up about front-end development. If you write front-end, this isn't about you personally. It's about how your choices make me angry. Also this is about how my choices have made me angry. Also this is mostly just about choices, the technologies are incidental. Note: The views expressedRead More

Most backcountry lightning education—including what’s taught by major outdoor organizations—is antiquated. Here’s your overdue expert update.