Will Oremus on why LLMs suffer from the “it’s not x, but Y” tic:

Even if researchers could figure out exactly why chatbots embrace negative parallelism, there’s another factor that could make it very hard to fix: “When something gets into these models, it’s very hard to pull it out,” Masrour, the Pangram engineer, said. That’s because one of the main ways that AI models have continued to evolve is by training on text generated by other bots. That AI text is presumably replete with negative parallelism, which further bakes it into the newer model. Now consider that a growing share of the writing on the internet is also AI-generated. This, too, becomes training data for future generations of AI.

Andrew McAfee on why he didn’t sign the open letter exhorting economists, policymakers, and technology leaders to act now to mitigate the effects of AI:

… It could bring risks, including large-scale job displacement, as well as opportunities such as major gains in living standards.

Dammit.

Why lead with the bad stuff here? Doesn’t AI already have enough of a ++PR problem++? Aren’t politicians from both sides of the aisle, in a rare show of bipartisanship, ++increasingly targeting AI++? Is coverage of AI in elite publications like The New York Times not ++negative enough++ for the authors of “We Must Act Now?”

As a group, economists strongly believe that technological progress, like free trade, is a good thing (both clearly raise standards of living, for example). But I heard Alan Blinder ruefully admit several years ago that economists had lost the public debate about the benefits of free trade. Do they really want to lose the debate about technological progress too? Do they want to contribute to that loss?

Good to know I’m not eating a credit card’s worth of plastic every week.

In an interview with Yale Environment 360, Rauert describes how, after learning about the potential for microplastics to infiltrate a lab — whether from construction materials, equipment, clothing, or the air — she and her fellow researchers painstakingly rebuilt their workspace, using glass and steel, to drastically reduce the odds of contamination. The hope is that once researchers can accurately measure levels of microplastics in human tissue and blood, they’ll be able to determine what exactly plastic pollution is doing to us.

“I don’t think we’ve got really good evidence at all for what effects [microplastics] might be having,” Rauert says. And the much-hyped finding that we eat a credit card’s worth of plastic each week? “That has absolutely been debunked.”

Good quote:

“It is inevitable that we face problems, but no particular problem is inevitable. We survive, and thrive, by solving each problem as it comes up. And, since the human ability to transform nature is limited only by the laws of physics, none of the endless stream of problems will ever constitute an impassable barrier. So a complementary and equally important truth about people and the physical world is that problems are soluble. By ‘soluble’ I mean that the right knowledge would solve them. It is not, of course, that we can possess knowledge just by wishing for it; but it is in principle accessible to”

― David Deutsch, The Beginning of Infinity: Explanations That Transform the World.

Academic food fight on what the hell ‘ultra-processed food” means:

The Science analysis, though, finds that the ultra-processed and non-ultra-processed diets in Hall’s study and other RCTs still differed in ways that could confound the trials’ results. The UPFs used in the trials tended to be softer than the non-ultra-processed foods, which can accelerate eating rate and lead to excess calorie consumption. They also tended to be more energy-dense (meaning that they had more calories per gram of food), less fiber-dense, and higher in saturated fat and sodium.

These aren’t inherent properties of “ultra-processed” foods. “While UPFs are often characterized by properties such as soft texture, higher energy density, and lower fiber content, these features are neither unique to UPFs nor consistently present across them,” Faidon Magkos, a professor in obesity and metabolism at the University of Copenhagen and an author of the Science analysis, told me in an email. “There are many non-UPF foods with similar properties, and conversely, many UPFs that are harder in texture, lower in energy density, or higher in fiber.” (To name one example, I eat these ultra-processed bars, which are hard in texture and a good source of fiber, multiple times a week.)