Feedback New scientist a popular sideways look at the latest science and technology news. You can submit content that you think might entertain readers in the Feedback section by sending an email. [email protected]
Speaking our truth
A seasoned science journalist soon learns to skim certain sections of scientific articles: in particular, sentences that say the research represents “significant progress” and “advances our understanding.” Not because they are wrong, but because literally any study that achieves anything at all can make such claims, and scientists have an incentive (like the rest of us) to amplify the impact of their work.
Except when they don't bother. In a long series of events that began with reporter Matthew Sparks and traveled through the social network Bluesky, Feedback discovered a 2018 paper on the arXiv preprint server that was poised to win a prize for its publication. absolute refusal make any big claims.
In it, researchers Joseph Redmon and Ali Farhadi described their latest version of YOLO: one of those artificial intelligence systems that can be trained to recognize objects in images. YOLO can pass CAPTCHA tests which ask you to click on all the bicycle squares and they were used to detect smuggling ships. This is all very impressive/worrying (remove if necessary), but by 2018 the pair had clearly started to decline.
It begins with the title of the article: “YOLOv3: Incremental Improvement.” The brief continues this trend, stating, “We've made a few small changes to the design to make it better.” The main text begins: “Sometimes you just call it a year, you know? I didn't do a lot of research this year. Spent a lot of time on Twitter.” That last line definitely dates the article.
The authors go on to explain that to improve YOLO, they “mostly took good ideas from other people.” They describe it in some detail, preliminarily admitting that the changes are “honestly nothing super exciting, just a bunch of little changes that make it better.”
We then move on to Section 4, which is called “Things We Tried That Didn't Work.” Feedback believes that this should be included in all scientific articles as a matter of course. This would save other researchers a lot of time.
The authors admit that they only described “what we can remember,” but remember that they tried to add something called “focus loss,” which made the model less accurate. “YOLOv3 may already be robust to the problem that focus loss is trying to solve,” they say, “because it has separate objectivity predictions and conditional class predictions. So for most examples there is no loss due to class predictions? Or something like that? We're not entirely sure.”
Reviews can't believe we missed this in 2018 or when we found out about it on Reddit aggregator site in 2024. But we are grateful to sociologist Per Engzell, who said on Bluesky that “Restrictions Sections where scientists practice radical honesty in exactly one paragraph,” and to data scientist Johan Ugander, who responded that the YOLOv3 paper should win an award for “the most honest newspaper“.
Surely someone must know about a scientist who is even more disarmingly honest about how little he has accomplished. Letters to regular address.
Long-lived bit
“I know you eschew nominative determinism,” Claire Boyce erroneously writes, “but I couldn’t resist sending you this email from the British Wildlife Newsletter today.” It was a book called Tree hunt: 1,000 trees to be found in cities across the UK and Ireland Paul Wood.
Likewise, Robert Masta notes that our recent How to Live to 100 (TL;DR Don't Die) special featured a longevity researcher named Paul Lazarus.
Sleep on it
Back in the distant past (July), Feedback wrote about receiving a press release in which strongly advocated for the sustainability of avocadosonly to discover that it came from the World Avocado Organization. We concluded that these people could be right or wrong, but either way they could be operating within the incentive structure.
We heard nothing more from the avocado market, but we did receive a series of press releases about the importance of sleep. “Can't find a solution? Science confirms that sleep really does solve problems,” said the first. He went on to share “exciting new research” explaining that “the old advice to 'sleep with it' may actually be one of the smartest problem-solving tools we have.”
This happens because the brain continues to process memories and form new connections while we sleep, he explains, sometimes generating new ideas by combining new and old ideas. They talk about “memory consolidation,” “the prefrontal cortex (the brain’s internal critic),” and “associative thinking.”
The follow-up email went even further, with a dramatic and grammatical headline: “New study shows rising mortality among young people, and their situation could get worse if sleep deprivation continues, experts warn.” The press release links poor sleep to chronic disease. There was also a quote from a “Certified Sleep Coach,” which may well have been a real thing, but in our confused minds it created an image of a sweaty man in a tracksuit blowing a whistle and yelling at us to “give me seven.” [hours]!” However, the message was clear: sleep well.
Perhaps the foreshadowing at the beginning gave it away, but in case you hadn't guessed, both letters were sent on behalf of Amerisleep, which is, of course, a mattress supplier.
Have a story for feedback?
You can submit stories to Feedback by email at [email protected]. Please provide your home address. Reviews for this and previous weeks may be saw on our website.
