Feedback is New Scientist A popular source for those seeking the latest in science and technology news. To share your thoughts on topics that may interest our readers, please send an email to feedback@newscientist.com.
Speaking Our Truth
Expert science journalists often learn to skim specific sections of scientific papers, particularly those asserting that the research signifies a “significant advancement” or “broadens understanding.” This isn’t because they are inaccurate, but because nearly any research yielding results can make these assertions, and academics are motivated (as we all are) to amplify the significance of their work.
But sometimes, it isn’t a hassle. Following a chain of events initiated by reporter Matthew Sparks and shared on the social platform Bluesky, Feedback uncovered a 2018 paper on the arXiv preprint server that would have claimed the prize for “most honest” research effort. Absolute refusal to make grand claims.
In this study, authors Joseph Redmon and Ali Farhadi detailed the latest iteration of YOLO, an AI framework designed to recognize objects in images. YOLO has even outperformed CAPTCHA tests requiring users to identify all squares containing bicycles and has been utilized to detect smuggling vessels. All said, these achievements are truly impressive (though perhaps alarming), but by 2018, the duo had evidently been coasting.
The title of their paper itself—”YOLOv3: Incremental Improvements”—is telling. The brief summary echoes this sentiment, stating, “we’ve implemented several minor design adjustments to enhance performance.” The essence can be distilled to: ‘You’ve contacted me year-round, yet my research efforts this year were sparse; much of my time was spent on Twitter.’ This line is, in fact, the date noted in the paper.
The authors further admit that the “improvements” largely stemmed from “good ideas taken from others.” They dive into details, first confessing that their tweaks are “not particularly exciting—just a collection of minor updates to enhance functionality.”
They then transition to Section 4, titled “Things I Tried That Didn’t Work.” This section, Feedback argues, should be a standard inclusion in all scientific publications, potentially saving others considerable time.
Despite acknowledging that they only recounted “what we recall,” they do remember an attempt involving something called “focal loss,” which ended up diminishing the model’s accuracy. “YOLOv3 might already be resilient to the focal loss problem, as it differentiates objectness prediction and conditional class prediction, resulting in minimal accuracy loss in most cases. Or maybe not? I’m not entirely sure,” they commented.
Feedback: I must have overlooked this in 2018, or I can’t believe I missed it when the article was spotlighted. Aggregator site Reddit played a role, but thanks to sociologist Per Angel, who mentioned in Bluesky:the restrictions section is a space for academics to practice radical honesty in just one paragraph. Data scientist Johan Ugander remarked that the YOLOv3 paper “deserves an accolade.” A truly candid piece.
Surely, there’s an academic somewhere known for their radical honesty regarding their minimal accomplishments. I’ll send an email to the usual address.
A Touch of Longevity
Clare Boyes once stated: “I understand you steer clear of prescriptive determinism, but I felt compelled to forward you this email I received today from the British Wildlife Newsletter.” It mentioned a book titled Tree Hunting: 1,000 Trees to Find in Towns and Cities Across the UK and Ireland, authored by Paul Wood.
In a similar vein, Robert Master pointed out that a recent special issue titled “How to Live to 100” (TL;DR: Don’t Die) featured a longevity researcher named Paul Lazarus.
Sleep on This
A while back (in July), Feedback reported on receiving a press release staunchly defending the environmental sustainability of avocados, only to find it originated from the World Avocado Organization. We concluded that while these individuals may be correct, they surely operate under a peculiar incentive structure.
We received no additional information from the avocado vendor, but we were inundated with press releases emphasizing the significance of sleep. “Struggling to find a solution? Science confirms that sleeping on it genuinely resolves your issues,” the first message proclaimed. It highlighted “exciting new research” and asserted that “the traditional advice to sleep on things may actually be one of the most effective problem-solving strategies available.”
This is attributed to the brain’s ability to continue processing memories and forging new connections while we sleep, occasionally leading to innovative insights through the amalgamation of new and old concepts. There was talk of memory consolidation, the prefrontal cortex (often seen as the brain’s inner critic), and associative thinking.
The follow-up email delved even further with a dramatic, albeit grammatically questionable, title: “New Study Indicates Rising Mortality Among Young Adults, Experts Warn Continued Sleep Deprivation Could Aggravate This.” The press release linked sleep deprivation with chronic health issues. There was also a quote from a “certified sleep coach”—possibly real?—but our minds conjured an image of a sweaty man in a tracksuit, whistle in hand, shouting, “Give me seven [hours]!” Yet the message remained clear: “Prioritize sleep.”
While it may have been foreshadowing, if you weren’t prepared for it, both emails, of course, came from mattress supplier Amerisleep.
Have a story for Feedback?
You can send your article to Feedback at feedback@newscientist.com. Please include your home address. This week’s and previous feedback can be accessed on our website.
Source: www.newscientist.com
