Correlations in Dias’s publication of an e-mail to Nature about manganese disulfide (MnS2)
After receiving an e-mail last year expressing concern about possible data fabrication in Dias’s PRL paper — a study not about room-temperature superconductivity, but about the electrical properties of manganese disulfide (MnS2) — the journal commissioned an investigation by four independent referees. Nature obtained documents about the investigation from sources who have asked to remain anonymous. The Editors of PRL wrote in the e-mail that the findings supported the allegations of data fabrication. Jessica Thomas, an executive editor at the American Physical Society, which publishes PRL, declined to comment.
Peter Armitage, a physicist at Johns Hopkins University in Baltimore, Maryland, who has been monitoring the controversy, says: “I just cannot see how we can trust anything [from Dias and Salamat] at this point.”
According to Ziemelis, the paper that was published this year was not influenced by the 2022 retraction of Dias research. “Our editors make decisions [about accepting manuscripts] based solely on whether research meets our criteria for publication,” he says. We will always look into concerns raised with us.
An analysis of the electrical resistance of germanium tetraselenide (GeSe4), MnS2 and other large language models
Hamlin noticed that the plot of the electrical resistance for the material germanium tetraselenide (GeSe4), discussed in the thesis was very similar to a plot of the figures in recent papers by Dias. Both plots had an extremely similar curve, especially at low temperatures, he says (see ‘Odd similarity’). It was very hard to imagine that this could be a coincidence.
After analysing the data, two of the four investigating referees concluded that the “only explanation of the similarity” in the GeSe4 and MnS2 plots is that data were taken from Dias’s 2013 thesis and used in the 2021 PRL paper. The referee found a mathematical function that can be used to map GeSe4 data onto the MnS2 data, which they demonstrated in their own report.
The race is on for new ways to assess ChatGPT and other large language models. eLife has a new peer-review model that depends on controversial claims about ancient humans.
The elephant in the room: the drongo bird of the African cuckoo (Cuculus gularis) isn’t an intelligent Geiger counter
Fork-tailed drongo birds (Dicrurus adsimilis) can spot the remarkably similar eggs of the parasitic African cuckoo (Cuculus gularis) with 94% accuracy. Each female drongo lays unique eggs that vary in colour and pattern from other females’. Individual females can only produce one kind of egg appearance because of the full range of this rich variety. They usually don’t drop a perfectly-matched egg into the nest, so drongos aren’t fooled. Even when researchers hand-picked a near-perfect interloper, drongos still kicked it out two-thirds of the time. Would you? In the image above, the cuckoo eggs are the ones on the lower right.
Archaeologists wowed viewers of a documentary — released last week — with stunning scenes of a cave crammed with bone fossils that, they argue, are the remains of the earliest-known burial by humans or their extinct relative Homo naledi. Four scientists who peer-reviewed a paper made the claims that the evidence was insufficient. The study is a high-profile test of the journal’s innovative publishing model: it no longer formally accepts papers, but instead publishes them alongside peer reviewers’ reports.
Scientists disagree on whether artificial intelligence can reason and are struggling with how to assess its capabilities. GPT-4 and other large language models ace academic exams and would probably breeze through a conversation-based Turing test. At the same time, they have glaring blind spots, such as failing an exam question when it is slightly changed, and they seem to lack abstract reasoning abilities. “There’s no Geiger counter we can point at something and say ‘beep beep beep — yes, intelligent’,” says cognitive scientist Tomer Ullman.
Andrew Robinson picks a book for the week that deals with the risk of a global pandemic, as well as a discussion on drug use by scientists.
Source: https://www.nature.com/articles/d41586-023-02439-2
Understanding the Deep-Sea Corals and Oceanoctopus Reefs: How Much Do We Know About Their Habitable Environments?
A lot of kids have unsafe levels of lead in their bloodstream. A ban on the manufacture and sale of lead paints is one approach that could eliminate one major source of exposure, according to the Lead Exposure Elimination Project (LEEP). A few thousand dollars can help get evidence of lead levels in a country’s paint. For example, a LEEP study in Pakistan led to ten paint manufacturers, including five of the most popular brands, switching to non-lead paint.
Beth Orcutt, an oceanographer, discovered a deep-sea octopus nursery last month and believes the world has not learned enough about deep-sea ecosystems to protect them. Our knowledge gaps are immense: deep-sea corals are keystone species, yet scientists don’t really know how they reproduce. And we have no idea whether it’s possible to restore damaged deep sea floor. “We need research covering at least ten years for each habitat to be able to make evidence-based decisions,” says Orcutt.