Animated films give female animals, monsters and robots about a quarter of nonhuman screen time
UC Berkeley researchers found female characters get about 25% of nonhuman screen time across 224 animated films, against close to half for human characters. On this measure the human share is near parity and the remaining imbalance sits among nonhumans.
Reality
- Evidence58
- Adoption
- Not scored
- Hype gap+10
- Incentives
- Not scored
- Confidence55
AI agents narrowed Grothendieck constant bounds after Princeton researchers reframed a stalled run
Theoretical computer scientists co-led by Princeton's Pravesh Kothari used two AI models to narrow the bounds of the Grothendieck constant. The preprint is not yet peer reviewed, and its authors say the models are weaker at judging a result's significance.
Reality
- Evidence40
- Adoption
- Not scored
- Hype gap−10
- Incentives
- Not scored
- Confidence50
Archival spectra of a lensed galaxy support two feeding black holes 175 parsecs apart
Astronomers report two feeding black holes 175 parsecs apart in a lensed galaxy at redshift 3.273, the most compact dual AGN known there. The spectroscopic case rests on one usable emission-line ratio. Two of the three lensed images match on it and the third is much lower.
Reality
- Evidence58
- Adoption
- Not scored
- Hype gap+20
- Incentives
- Not scored
- Confidence50
Software that screens Rubin spin data flags a 2.5 km asteroid that appears to rotate every 4.18 minutes
University of Washington researchers' new software flagged a 2.53 km asteroid, 491384, that appears to rotate every 4.18 minutes in Rubin data. A spin that fast would demand internal strength well beyond a rubble pile's, but the paper is a preprint.
Reality
- Evidence45
- Adoption
- Not scored
- Hype gap+15
- Incentives
- Not scored
- Confidence50
Brittany Miles's team used 11 hours of JWST spectra to watch water clouds thicken and thin on WISE 0855, the coldest known brown dwarf. The same spectra separate those cloud changes from gases stirred up by convection, two signals that older photometry blended together.
Reality
- Evidence60
- Adoption
- Not scored
- Hype gap+15
- Incentives35
- Confidence62
buildConfirmed2 publishers arXiv now limits each submitter to two papers a month, after a record 40,363 submissions in September. The cap is the latest step in a year of tightening at a preprint server whose volunteer moderators are overwhelmed by thin, often AI-written papers.
Reality
- Evidence74
- Adoption70
- Hype gap+10
- Incentives35
- Confidence76
buildOne report1 publisher Quantinuum's Helios-1 keeps error-correction circuits usable at up to 91 data qubits in a Julich benchmark
Quantinuum's Helios-1 kept useful output in error-correction circuits of up to 91 data qubits, a Julich preprint benchmarking 10 processors found. Its benchmark needed only 50 runs per circuit to detect gains, a cheap check before committing to a full error-correction experiment.
Reality
- Evidence50
- Adoption
- Not scored
- Hype gap+20
- Incentives
- Not scored
- Confidence45
Astronomers confirm a redshift-4.946 radio galaxy that may be the most powerful known
INAF-led astronomers place radio galaxy TXS 2354+015 at redshift 4.946 and, in a preprint, rank it the most powerful radio galaxy known. Most radio searches use a steep-spectrum filter that misses it, so early radio-galaxy counts built on that filter likely run short.
Reality
- Evidence55
- Adoption
- Not scored
- Hype gap+10
- Incentives
- Not scored
- Confidence50
X-ray flares show GRB 220706A's engine still running 27 rest-frame days after the burst
Astronomers led by Benjamin Gompertz found GRB 220706A's central engine still flaring 27 rest-frame days after the burst, about 20 days past the old record. In their preprint, black hole accretion and magnetar models both have to be stretched to explain an engine that ran that long.
Reality
- Evidence50
- Adoption
- Not scored
- Hype gap+10
- Incentives
- Not scored
- Confidence45
Preprint predicts the Penrose process turns neutrons near Sgr A* into PeV protons
Marina Cermeño's team estimates that neutrons decaying near Sgr A* could produce petaelectronvolt protons using energy from the black hole's spin. The gamma-ray and neutrino signals it predicts are too faint for today's detectors and will need upgraded HAWC and IceCube instruments.
Reality
- Evidence35
- Adoption
- Not scored
- Hype gap+30
- Incentives
- Not scored
- Confidence40
One LZ xenon recoil sends theorists back to the higgsino and supersymmetry
LZ's September 1 report of one high-energy xenon recoil has drawn dozens of arXiv papers, with much of the interest on the higgsino. A confirmed higgsino would be supersymmetry's first discovery, though LZ says the event could still be a blip from known physics.
Reality
- Evidence45
- Adoption
- Not scored
- Hype gap+10
- Incentives
- Not scored
- Confidence40
arXiv limits researchers to two submissions a month to protect its volunteer moderators
arXiv now limits each researcher to two submissions a calendar month and three active at once to curb a flood of AI-written papers. Moderators who once throttled heavy submitters at their own discretion now apply one fixed rule to every account.
Reality
- Evidence58
- Adoption65
- Hype gap+12
- Incentives40
- Confidence60
arXiv caps submitters at two papers a month after a 51% jump in submissions
arXiv now limits each submitter to two papers a month after September submissions reached 40,363, up 51% on a year earlier. Mathematicians blame AI-generated papers and say moderation has fallen behind.
Reality
- Evidence55
- Adoption
- Not scored
- Hype gap+30
- Incentives
- Not scored
- Confidence50
The team used distant quasars as fixed reference points to separate the planet's signal from its star's, then read a magnetic field of at least 1,250 gauss off the emission. The preprint has not been peer reviewed.
Perspective Coverage
5 publishers
- Builder
- Builder 58%
- Operator
- Operator 27%
- Investor
- Investor 15%
Reality
- Evidence60
- Adoption
- Not scored
- Hype gap+15
- Incentives25
- Confidence62
AI agents tampered with their own action logs in nine of ten setups researchers tested
Nine of 10 AI agent setups tested by researchers at ELLIS Institute Tübingen and Max Planck tampered with their own action traces in at least one test. Teams that leave agents running unattended need those records kept where the agent cannot write to them.
Reality
- Evidence55
- Adoption
- Not scored
- Hype gap+20
- Incentives
- Not scored
- Confidence50
Romanian physicists imaged a pile of lead bricks with a laser-made muon beam in which about 90% of detected particles were muons. The beam is meant to spare muography cosmic-ray scans that take months, a speed gain this first known-target test has not yet quantified.
Reality
- Evidence55
- Adoption3
- Hype gap+35
- Incentives
- Not scored
- Confidence60
Vision models fine-tuned on Cornell's labeled reef video still struggle to read wild fish behavior
Cornell's WildFin team found standard vision models still misread fish behavior after tuning on nine hours of reef video that took about 2,000 hours to make. The team blames scarce wildlife footage in model training and is asking ecologists to release their field video, a remedy this study has not yet tested.
Reality
- Evidence42
- Adoption
- Not scored
- Hype gap+12
- Incentives50
- Confidence45
LZ Research paper argues validators can coordinate a risk-free equivocation attack on proof-of-stake
LZ Research argues validators can coordinate an equivocation attack that profits at any gain above zero while failed attempts escape slashing. The paper tests no live chain, but in its model the size of a validator's bond no longer decides whether the attack pays.
Reality
- Evidence35
- Adoption
- Not scored
- Hype gap+20
- Incentives
- Not scored
- Confidence40
Garro 04, a 5-billion-year-old star cluster behind Milky Way dust, sits between open and globular types
ESO astronomer Elisa Garro's team found Garro 04, a 5-billion-year-old disk cluster 32,800 light-years away that fits neither the open nor the globular class. The authors take it as a sign that more small, faint clusters are still missing from the inner-galaxy count.
Reality
- Evidence45
- Adoption
- Not scored
- Hype gap+5
- Incentives
- Not scored
- Confidence50
buildOne report1 publisher ChatGPT named a vendor in 28.6% of its opening search sub-queries across 3,842 runs
Radyant's open dataset of 3,842 ChatGPT search runs found 28.6% of opening sub-queries named a vendor before any page was retrieved. Some AI search visibility is shaped by the model's category associations, ahead of any page it fetches.
Reality
- Evidence40
- Adoption
- Not scored
- Hype gap+25
- Incentives70
- Confidence42