SORT
CATEGORY
Good
P(generalises) × big | true
0.8 × 7 = 5.6

Self-driving cars are roughly 10x safer than human drivers in the same locations

+
n=96 million miles on the road in 4 US cities. No highway data. Self-reported data but with some legal force against nonreporting. This lines up with an earlier, mostly-third-party study by Swiss Re using postcode matching.
P(generalises) = 0.8
Big | True = 7
Big * P(generalises) = 5.6
AI
Infrastructure
Mixed
P(generalises) × big | true
0.8 × 6.9 = 5.52

It's quite easy to accidentally undo current AI alignment methods, e.g. by just training on some naughty numbers

+
Emergent misalignment is the following startling fact: 1. take a modern “aligned” assistant LLM; 2. Train it to write intentionally insecure code; 3. It starts praising Hitler. The Anthropic followup study found that this also happens with more aligned models, in real production environments, and could easily happen accidentally. Actually there's an upside, which is that if bad things are correlated then we can push against them all at once as a bundle.
P(generalises) = 0.8
Big | True = 6.9
Big * P(generalises) = 5.52
AI
Safety
Good
P(generalises) × big | true
0.9 × 6 = 5.4

AI systems discovered many critical "zero-day" vulnerabilities in core code like OpenSSL - and also provided a patch

+
Google’s Big Sleep agent and the startup AISLE found dozens of critical vulnerabilities in some of the main infrastructure of the internet: Linux, cURL, OpenSSL, and SQLite.
P(generalises) = 0.9
Big | True = 6
Big * P(generalises) = 5.4
AI
Cybersecurity
Bad
P(generalises) × big | true
0.65 × 8 = 5.2

Artificial chimeras of bat coronaviruses, with up to 100% lethality, produced, on purpose, under BSL-2 conditions

+
Intentional design of chimera viruses which transfer across species (pig to mouse). n=83. Conducted in BSL-2 conditions, grossly inadequate. Guangzhou Medical University study published in the US Journal of Virology.
P(generalises) = 0.65
Big | True = 8
Big * P(generalises) = 5.2
Biology
Virology
Genetic engineering
Good
P(generalises) × big | true
0.9 × 5.75 = 5.175

The oldest baby in history: an embryo frozen in 1994 was brought to term and resulted in a healthy baby boy

+
His mother Lindsey is in some sense only 5 years older than him. The previous record was 25 years. This is a case of a successful pregnancy with a donated embryo. The biological mother of the baby was 62 years old at the time of his birth. While incremental, this points towards eventually allowing for delayed IVF, which would be socially transformative.
P(generalises) = 0.9
Big | True = 5.75
Big * P(generalises) = 5.175
Biology
Medicine
Cryonics
Good
P(generalises) × big | true
0.85 × 6 = 5.1

Whole-population follow-up on the Covid vaccines shows a large decrease in all-cause mortality

+
Average 4 year followup. In France (n=29 million) the vaccinated had a 25% decrease in all-cause mortality (and 74% less risk of Covid death). A Norway study (n=5m) found extremely similar results.
P(generalises) = 0.85
Big | True = 6
Big * P(generalises) = 5.1
Biology
Public health
Covid
Good
P(generalises) × big | true
0.9 × 5.5 = 4.95

AIs contributed to solving and formalising many open problems in research mathematics

+
Most of the results are improving on known bounds in number theory or optimisation, with a lively community also using them on the Erdos Problems (which also include graph theory, ramsey theory, combinatorics, and geometry). In most instances, humans provided the whole proof strategy and thousands of tokens of steering and some final cleanup. It also remains true that AIs don't yet come up with novel concepts, just use "slight variant[s] of standard facts".
P(generalises) = 0.9
Big | True = 5.5
Big * P(generalises) = 4.95
AI
Mathematics
Good
P(generalises) × big | true
0.7 × 7 = 4.9

The claimed link between fluoride exposure and lower IQ doesn't hold up

+
Critical study of Taylor et al., 2025, which linked fluoride exposure with lowered IQ. Main criticisms: bias/weakness of included studies, inconsistency across studies (e.g., what counts as high vs low fluoride), faulty stats.
P(generalises) = 0.7
Big | True = 7
Big * P(generalises) = 4.9
Biology
Public health
Toxicology
Mixed
P(generalises) × big | true
0.7 × 7 = 4.9

The first detection of extraterrestrial glucose, and also the last missing ingredient for RNA (ribose)

+
Found on the first try! From asteroid Bennu, brought back to Earth by the OSIRIS-REX mission, the first ever mission to bring back a significant amount of material. Bennu thus had all of the basic ingredients for life: the nucleotides, amino acids, and sugars. This suggests they are common. I put this as Big If True because of the implications for the 'Great Filter' debate: if abiogenesis is easy then something else has to explain the Fermi paradox.
P(generalises) = 0.7
Big | True = 7
Big * P(generalises) = 4.9
Astronomy
Biochemistry
Good
P(generalises) × big | true
0.7 × 7 = 4.9

Diagnostics on a phone with no doctor needed

+
P(generalises) = 0.7
Big | True = 7
Big * P(generalises) = 4.9
AI
Medicine
Good
P(generalises) × big | true
0.85 × 5.75 = 4.89

Approval of a strong non-opioid painkiller targeting a pathway specific to pain neurons

+
Suzetrigine, in the US. It’s a sodium-channel inhibitor. Importantly, unlike earlier such drugs like lidocaine, it's specific to the NaV1.8 sodium channels expressed in pain-sensing neurons. It's strong, as strong as hydrocodone or low-dose morphine. It is claimed to be nonaddictive, which is the somewhat unlikely part. It does hit the brain much less, which might work. Not approved for chronic pain, only acute.
P(generalises) = 0.85
Big | True = 5.75
Big * P(generalises) = 4.89
Biology
Medicine
Painkillers
Good
P(generalises) × big | true
0.97 × 5 = 4.85

A new class of treatment for malaria: 97% cured and it shouldn't suffer existing drug resistances

+
GanLum. The previous best treatment was already 94% effective. "the first new class of malaria drug approved in more than 25 years.". See also the identification of a new family of natural antibodies.
P(generalises) = 0.97
Big | True = 5
Big * P(generalises) = 4.85
Biology
Medicine
Malaria treatments
Mixed
P(generalises) × big | true
0.8 × 6 = 4.8

Unreleased AIs won gold at three of the hardest cognitive competitions in the world: IMO, ICPC, and IOI

+
The AI submissions respected the human time cap (9 hours of clock time). Reportedly both the OpenAI and Deepmind systems are just an LLM (i.e. no external tools). They probably cost an insane amount to run, but for our purposes this is fine: we want the capability ceiling rather than the productisable ceiling.
P(generalises) = 0.8
Big | True = 6
Big * P(generalises) = 4.8
AI
Mathematics
Good
P(generalises) × big | true
0.8 × 6 = 4.8

Synthesis of hexanitrogen, the most energy-dense molecule ever

+
The process works at room temperature, but the product breaks down after 36 milliseconds - but that’s long enough to cool it normally and so keep it stable indefinitely. “it would be the most useful rocket fuel on the planet. N6 would not burn with a flame: it’s just a burst of energy that generates a large volume [of gas] – so a lot of thrust – and it is non-corrosive”.
P(generalises) = 0.8
Big | True = 6
Big * P(generalises) = 4.8
Chemistry
Energy
Bad
P(generalises) × big | true
0.6 × 8 = 4.8

The 83% cut to USAID is on track to cause millions of additional deaths over the next 5 years

+
However, data quality is systematically terrible in the developing world. Just a regression study without much of any causal identification. The original result was 8 to 20 million excess deaths, but this is fairly unlikely. But a sceptical replication (below) still predicts around 1-4m deaths (based on the questionably basic Regression 1 analysis). There is a patchwork of new bilateral agreements planned, which might reduce the damage by a factor of 5.
P(generalises) = 0.6
Big | True = 8
Big * P(generalises) = 4.8
Social
Bad
P(generalises) × big | true
0.6 × 8 = 4.8

Proposed 40% cut to the NIH estimated to kill 50% of new treatments

+
The proposed cuts to American science are unprecedented, but it's not too late. The world is about to learn what it’s like to fly without the engine. The model assumes that the 40% lowest-priority grants are the ones lost, which is not a foregone conclusion. This implies mild anticorrelation between the NIH's scoring and impact.
P(generalises) = 0.6
Big | True = 8
Big * P(generalises) = 4.8
Social
Good
P(generalises) × big | true
0.7 × 6.8 = 4.76

Extreme poverty drops from 27% of India to 5% in one decade

+
P(generalises) = 0.7
Big | True = 6.8
Big * P(generalises) = 4.76
Social
Mixed
P(generalises) × big | true
0.65 × 7 = 4.55

You can now program human neurons entirely over an API

+
The ominous-sounding FinalSpark's Neuroplatform is $1000/month and handles all of the organoid growth and chemical machinery for you. From October 2024. Brains are many thousands of times more power-efficient than current silicon, so this could change the world. Ethical questions remain.
P(generalises) = 0.65
Big | True = 7
Big * P(generalises) = 4.55
Biology
Computing
Neuroscience
Good
P(generalises) × big | true
0.65 × 7 = 4.55

Murder rates worldwide have fallen 25% since 2000

+
On average! Potentially some confounding from improved trauma emergency care converting murders into attempted murders.
P(generalises) = 0.65
Big | True = 7
Big * P(generalises) = 4.55
Social
Good
P(generalises) × big | true
0.9 × 5 = 4.5

We produce antimatter eight times faster than last year

+
"eight-fold increase in the trapping rate of antihydrogen, enabled bysympathetic cooling of positrons with laser-cooled beryllium ions." Still only 2000 atoms an hour.
P(generalises) = 0.9
Big | True = 5
Big * P(generalises) = 4.5
Physics
Engineering
Bad
P(generalises) × big | true
0.9 × 5 = 4.5

Materials scientists warn of threat posed by AI-generated microscopy images

+
P(generalises) = 0.9
Big | True = 5
Big * P(generalises) = 4.5
AI
Materials science
Good
P(generalises) × big | true
0.9 × 5 = 4.5

Automated acoustic monitoring of bird species with full natural language understanding. Can ask it things like "how many babies are there in earshot?"

+
Perch is DeepMind's model of bioacoustic data. It's trained on a range of animal vocalizations (14k-15k species), from mammals and birds to underwater samples and anthropogenic noise. It's being integrated into widely-used tools like Cornell's BirdNet Analyzer and used in monitoring endangered species.
P(generalises) = 0.9
Big | True = 5
Big * P(generalises) = 4.5
Biology
Zoology
Animal communication
Good
P(generalises) × big | true
0.9 × 5 = 4.5

The cost to treat drug-resistant TB drops below $300

+
BPaLM regimen is the most widely recommended treatment for drug-resistant tuberculosis. A part of this treatment is bedaquiline produced by Lupin which is now available at $63 per treatment course, bringing the price of the complete BPaLM treatment below $300 (at $284).
P(generalises) = 0.9
Big | True = 5
Big * P(generalises) = 4.5
Social
Epidemiology
Infectious diseases
Bad
P(generalises) × big | true
0.5 × 9 = 4.5

The last universal common ancestor arose in "a surprisingly short interval of geologic time" and was surprisingly similar to modern bacteria

+
"It suggests that life must have got started on Earth pretty much as soon as it possibly could have done. Which in turn implies that, given the right conditions and ingredients, life might not be an extremely rare and unlikely accident, as some scientists have believed, but rather, almost an inevitability, and therefore likely to be abundant in the universe." With one more leap of logic, we can weakly infer that the Great Filter (the unknown thing which prevents intelligent life from sending signals and making themselves known to us) is more likely to be ahead of us and that this is terrible, terrible news.
P(generalises) = 0.5
Big | True = 9
Big * P(generalises) = 4.5
Biology
Evolutionary Biology
Good
P(generalises) × big | true
0.99 × 4.5 = 4.455

The first images of individual atoms interacting in a gas

+
More accurately: the first in-situ atom-by-atom imaging of itinerant atom pair correlations in a continuous 2D gas. Published simultaneously by three different labs (Yao MIT, de Jongh ENS, Xiang MIT). Prior microscopy took pictures of individual atoms, but only in heavily restricted optical lattices and tweezers.
P(generalises) = 0.99
Big | True = 4.5
Big * P(generalises) = 4.455
Physics
Engineering
Quantum gas microscopy
Good
P(generalises) × big | true
0.98 × 4.5 = 4.41

Four countries eliminated trachoma, a disease that causes blindness

+
This year Senegal, Egypt, Mauritania and Fiji joined the list of countries that have successfully eliminated trachoma as a public health problem. Trachoma is caused by the bacterium Chlamydia trachomatis which is highly infectious and spreads through bodily fuilds and direct contact.
P(generalises) = 0.98
Big | True = 4.5
Big * P(generalises) = 4.41
Social
Epidemiology
Infectious diseases
Good
P(generalises) × big | true
0.8 × 5.5 = 4.4

First human infant cured of a lethal genetic disease with a personalized gene therapy

+
Baby KJ was born with a rare mutation in a urea cycle enzyme, carbamoyl phosphate synthase. This mutation results in the build-up of ammonia and high early mortality. A team of researchers at Children's Hospital of Philadelphia developed a full treatment pipeline, including a specific base editor (CRISPR targeting one letter in the DNA) and mRNA packaged into lipid nanoparticles targeting the liver. The whole process took 6 months, and the baby is doing well.
P(generalises) = 0.8
Big | True = 5.5
Big * P(generalises) = 4.4
Biology
Medicine
Gene therapy
Good
P(generalises) × big | true
0.95 × 4.5 = 4.275

First control over a single antiproton

+
i.e. one particle precisely manipulated using microwave pulses. Resulted in the most precise measurement of its magnetic moment by a factor of 16.
P(generalises) = 0.95
Big | True = 4.5
Big * P(generalises) = 4.275
Physics
Spectroscopy
Good
P(generalises) × big | true
0.85 × 5 = 4.25

Three new countries certified malaria-free

+
Georgia, Suriname, and Timor-Leste have been certified malaria-free
P(generalises) = 0.85
Big | True = 5
Big * P(generalises) = 4.25
Social
Mixed
P(generalises) × big | true
0.85 × 5 = 4.25

In March, the Nucleic Acid Observatory identified their first engineered virus in wastewater

+
In this case, the virus and the edits were benign and there was no indication of this being an accidental leak. The process is heavily automated and their tools are open source.
P(generalises) = 0.85
Big | True = 5
Big * P(generalises) = 4.25
Biology
Engineered pathogens
Monitoring
Good
P(generalises) × big | true
0.65 × 6 = 4.2

First-in-human 'prime editing' gene therapy. Cured an inherited immune disease.

+
Clinical prime editing of CD34+ cells to treat the immune disease p47-CGD in n=2 patients. 6 month followup. PE is a CRISPR derivative that doesn't cut both strands of DNA and writes new code directly, allowing for much less off-target damage. It's a Cas9 nickase plus a guided reverse transcriptase (pegRNA). Fewer unwanted insertions/deletions, and opens up more edit types (substitutions, smaller indels). Just somatic edits, as usual. Big if true score reflects the method, not this condition (1 in 200,000).
P(generalises) = 0.65
Big | True = 6
Big * P(generalises) = 4.2
Biology
Medicine
Oncology
Mixed
P(generalises) × big | true
0.7 × 6 = 4.2

Maybe a majority of the Ukrainian drone arsenal is 3D-printed and metal printed

+
P(generalises) = 0.7
Big | True = 6
Big * P(generalises) = 4.2
Hardware
Engineering
Military
Good
P(generalises) × big | true
0.6 × 7 = 4.2

5 factors explain most of the genetic variance in common mental illnesses

+
n=1 million genomes and 14 mental illness phenotypes. The five factors explain 66% of the specifically genetic variance of the individual disorders. Schizophrenia and bipolar disorders and major depression, PTSD and anxiety overlap strongly.
P(generalises) = 0.6
Big | True = 7
Big * P(generalises) = 4.2
Biology
Medicine
Genetics
Good
P(generalises) × big | true
0.6 × 7 = 4.2

Large effect for 5-MeO-DMT for treatment-resistant depression

+
Phase 2b trial (n=193) supposedly yielding an effect size of d=1. Toad venom for what ails you. Recall that major depression is maybe 2% of the total global burden of disease.
P(generalises) = 0.6
Big | True = 7
Big * P(generalises) = 4.2
Biology
Medicine
Psychoactive drugs
Good
P(generalises) × big | true
0.6 × 7 = 4.2

For the first time in recent history, China’s emissions might be falling

+
P(generalises) = 0.6
Big | True = 7
Big * P(generalises) = 4.2
Social
Good
P(generalises) × big | true
0.6 × 7 = 4.2

The first evidence of a solar take-off in Africa

+
P(generalises) = 0.6
Big | True = 7
Big * P(generalises) = 4.2
Social
Good
P(generalises) × big | true
0.7 × 6 = 4.2

Two new Earthlike planets within 20 lightyears have been discovered, one of which is the first known to orbit a G-type star like ours

+
Their “habitability” is not really known, we’re just guessing by how far out they orbit their star.
P(generalises) = 0.7
Big | True = 6
Big * P(generalises) = 4.2
Physics
Astronomy
Good
P(generalises) × big | true
0.7 × 6 = 4.2

The world's first-ever thorium molten salt reactor is now operational

+
Prototype: only 2MW. The only operational molten salt reactor in the world to have successfully incorporated thorium fuel. The first operating MSR in 56 years.
P(generalises) = 0.7
Big | True = 6
Big * P(generalises) = 4.2
Hardware
Energy
Mixed
P(generalises) × big | true
0.7 × 6 = 4.2

AIs reliably associate arbitrary numbers with very particular concepts. For instance, 87 is reliably taken to imply "owl"

+
This "subliminal learning" effect only applies within a single model, so you only get spooky communication when it's two copies of a model communicating. And "87" is only "owl" for one specific checkpoint of GPT-4, so you'd have to rediscover the hidden dictionary each time.
P(generalises) = 0.7
Big | True = 6
Big * P(generalises) = 4.2
AI
Bad
P(generalises) × big | true
0.7 × 6 = 4.2

Claude outperforms nearly all human competitors in basic hacking competitions

+
Minimal human assistance and ordinary amounts of per-query compute.
P(generalises) = 0.7
Big | True = 6
Big * P(generalises) = 4.2
AI
Cybersecurity
Good
P(generalises) × big | true
0.7 × 6 = 4.2

The American divorce rate has roughly halved since 2000

+
P(generalises) = 0.7
Big | True = 6
Big * P(generalises) = 4.2
Social
Good
P(generalises) × big | true
0.7 × 6 = 4.2

Last year's biannual HIV shot available in low-income countries, $40/year

+
P(generalises) = 0.7
Big | True = 6
Big * P(generalises) = 4.2
Social
Bad
P(generalises) × big | true
0.7 × 6 = 4.2

Patents also have a replication crisis

+
P(generalises) = 0.7
Big | True = 6
Big * P(generalises) = 4.2
Social
Bad
P(generalises) × big | true
0.7 × 6 = 4.2

One in four social science papers is straightforwardly wrong

+
P(generalises) = 0.7
Big | True = 6
Big * P(generalises) = 4.2
Social
Metascience
Quality control
Good
P(generalises) × big | true
0.75 × 5.5 = 4.125

Trevogrumab could potentially prevent muscle loss in the sedentary

+
The unusually badly-named Trevogrumab invites us to speculate about a world in which everyone, no matter how lazy or elderly, is jacked. Wine pairing for Ozempic. n=599 phase 2 trial where the loss halved.
P(generalises) = 0.75
Big | True = 5.5
Big * P(generalises) = 4.125
Biology
Medicine
Body enhancement
Good
P(generalises) × big | true
0.9 × 4.5 = 4.05

Highest-res image of an atom / first direct observation of moiré phasons

+
"Real-space" thermal vibrations recorded from twisted bilayer WSe2 atoms via electron ptychography. The actual "image" is computationally reconstructed and complex-valued, so the colour image below is figurative. But less figurative than all others!
P(generalises) = 0.9
Big | True = 4.5
Big * P(generalises) = 4.05
Physics
Photography
Good
P(generalises) × big | true
0.9 × 4.5 = 4.05

The first superfluid molecule

+
That is, the zero-viscosity superfluid phase of matter was spotted for the first time in a molecule (H2 hydrogen). The state only occurs around 1 Kelvin, i.e. 10K below the freezing point of hydrogen. First predicted in 1972, previously observed in atomic helium.
P(generalises) = 0.9
Big | True = 4.5
Big * P(generalises) = 4.05
Physics
Quantum chemistry
Good
P(generalises) × big | true
0.9 × 4.5 = 4.05

The EU eases its ban on GMO food, allowing newer gene editing techniques

+
NGT Category 1 allows genetic editing up to some short length. This is actually the same as the US FDA arrangement.
P(generalises) = 0.9
Big | True = 4.5
Big * P(generalises) = 4.05
Social
Biology
Bad
P(generalises) × big | true
0.8 × 5 = 4

There exists a large black market in scientific fraud

+
P(generalises) = 0.8
Big | True = 5
Big * P(generalises) = 4
Social
Good
P(generalises) × big | true
0.8 × 5 = 4

The first practical muon beam

+
First laser-plasma accelerator, a compact source of high-flux, directional, penetrating muons. one foot long. Could be used to scan large objects (where radio and x-rays struggle). Yields are so much higher that the scan is thousands of times faster.
P(generalises) = 0.8
Big | True = 5
Big * P(generalises) = 4
Physics
Engineering
Bad
P(generalises) × big | true
0.8 × 5 = 4

Just-in-time AI malware: the first software viruses which use LLMs to self-modify

+
PROMPTSTEAL (effective and deployed) uses Hugging Face models to generate Windows commands, which lets it remove nearly all suspicious code from its payload. PROMPTFLUX (experimental, not yet effective) calls Gemini over API to alter its source code.
P(generalises) = 0.8
Big | True = 5
Big * P(generalises) = 4
AI
Cybersecurity
Good
P(generalises) × big | true
0.8 × 5 = 4

There's a pacemaker smaller than a grain of rice

+
For patients who need temporary pacing (to recover after a surgery, or while waiting for a permanent pacemaker). The temporary pacemaker can be injected through the skin and dissolves when no longer needed, as all its components are biodegradable. Paired with a skin-contact wireless device to detect arrhythmias. No wires, no batteries, no need for surgery to remove it from the body. So far shown to work in mouse, rat, pig, dog and human donor hearts.
P(generalises) = 0.8
Big | True = 5
Big * P(generalises) = 4
Biology
Medicine
Medical devices
Good
P(generalises) × big | true
0.8 × 5 = 4

Gene therapy for heritably high cholesterol

+
Verve-102 targets the PCSK9 gene to permanently lower low density lipoprotein (LDL-C) expression in the liver after a single infusion. A lipid nanoparticle carrying a guide RNA targeting PCSK9 and an RNA base editor enzyme. Tested in primary human hepatocytes, mice and non-human primates. Verve got FDA Fast Track designation & was bought by Lilly.
P(generalises) = 0.8
Big | True = 5
Big * P(generalises) = 4
Biology
Medicine
Gene therapy
Good
P(generalises) × big | true
0.8 × 5 = 4

First real test of mitochondrial replacement therapy for mtDNA Disease

+
First real test of mitochondrial replacement therapy IVF. n=8 healthy babies born to mothers with dangerous mitrochondrial homoplasmy genes (sparing the babies from sometimes-fatal diseases), worked so far in 100% of treated pregnancies. Up to 2 year followup, normal development so far.
P(generalises) = 0.8
Big | True = 5
Big * P(generalises) = 4
Biology
Medicine
Mitochondrial replacement therapy
Good
P(generalises) × big | true
0.8 × 5 = 4

The first rollout of a gonorrhoea vaccine

+
NHS rollout after a 2022 analysis noticed that repurposing the meningitis B vaccine works somewhat (30-40% effective on resistant cases). There's a recent surge in the incidence of gonorrhea in the UK. The repurposed MenB vaccine, though not highly efficient, is hoped to curb the spread of the gonorrhea that became resistant to antibiotic treatment.
P(generalises) = 0.8
Big | True = 5
Big * P(generalises) = 4
Biology
Medicine
Vaccines
Good
P(generalises) × big | true
0.8 × 5 = 4

The brain’s most abundant receptor works quite differently than textbooks assume

+
GluA2 AMPA glutamate receptors have been thought to be impermeable for calcium ions. This paper provides evidence to the contrary: Glu2A AMPA receptors in fact show a range of calcium permeability depending on the subunit composition of the receptor tetramers, spatial orientation of transmembrane regulatory proteins and auxiliary subunits that affect the receptor's selectivity filter. A major conceptual update in basic neurophysiology, with potential implications for autism, ALS, epilepsy etc. This may in part explain pathogenicity of missense mutations in AMPAR.
P(generalises) = 0.8
Big | True = 5
Big * P(generalises) = 4
Biology
Neuroscience
Neurotransmitter receptors
Good
P(generalises) × big | true
0.8 × 5 = 4

A new algorithm for computing the product of matrix by its transpose

+
P(generalises) = 0.8
Big | True = 5
Big * P(generalises) = 4
Mathematics
AI
Bad
P(generalises) × big | true
0.99 × 4 = 3.96

First pardon for a corporation, BitMEX

+
In the US.
P(generalises) = 0.99
Big | True = 4
Big * P(generalises) = 3.96
Social
Good
P(generalises) × big | true
0.99 × 4 = 3.96

Space is full of things: the 3rd-ever recorded interstellar object flew by

+
We've seen 3 in our solar system in just 8 years: 1I/ʻOumuamua, 2I/Borisov, and now 3I/ATLAS
P(generalises) = 0.99
Big | True = 4
Big * P(generalises) = 3.96
Physics
Astronomy
Good
P(generalises) × big | true
0.98 × 4 = 3.92

Progress on generalising modularity past elliptic curves

+
Every abelian surface in one major class can now be associated to a modular form
P(generalises) = 0.98
Big | True = 4
Big * P(generalises) = 3.92
Mathematics
Good
P(generalises) × big | true
0.98 × 4 = 3.92

A new algorithm for shortest-path finding, the first faster than a sort

+
Improvement to shortest-path finding, the first which is faster than a sort. Bellman’s revenge. The trick is to not require ordered outputs. It’s fiddly (56 lines of pseudocode, divide-and-conquer-Bellman-frontier-reduction), so unlikely to be used except in large graphs. No reason to think this new Duan algo is optimal either.
P(generalises) = 0.98
Big | True = 4
Big * P(generalises) = 3.92
Mathematics
CS
Good
P(generalises) × big | true
0.95 × 4 = 3.8

Long-awaited explanation for the mysterious success of the world's default linear programming algorithm

+
i.e. Simplex. A long-awaited theoretical explanation for why simplex completely avoids exponential slowdowns suggested by worst-case analysis. They also give a theoretical speedup for it, and a proof of their speedup's optimality in a certain sense (noise dependence up to polylog), but the constants are so large that this is unlikely to ever be used.
P(generalises) = 0.95
Big | True = 4
Big * P(generalises) = 3.8
Mathematics
Computer science
Complexity theory
Bad
P(generalises) × big | true
0.75 × 5 = 3.75

A global pandemic has apparently wiped out sea urchin populations

+
74% to 99.7% mortality in Diadema africanum in studied sites. Die-offs are not uncommon, but a second wave epidemic in 2023 appears to have disrupted their breeding. Urchins are crucial in controlling seaweed and algae blooms.
P(generalises) = 0.75
Big | True = 5
Big * P(generalises) = 3.75
Biology
Zoology
Marine biology
Good
P(generalises) × big | true
0.75 × 5 = 3.75

Blood cancer treatment nearly triples survival-without-progression, to 83%

+
Specifically for patients with relapsing multiple myeloma. Phase III trial (n-587), 4 year followup against an active and gold-standard control.
P(generalises) = 0.75
Big | True = 5
Big * P(generalises) = 3.75
Biology
Medicine
Oncology
Good
P(generalises) × big | true
0.8 × 4.5 = 3.6

The first 2D metal sheet

+
Actually two atoms thick, roughly 6 ångström. In bismuth, gallium, indium, tin and lead. Youtubers vindicated; the process involves a big hydraulic press (van der Waal squeezing). Previous methods didn't produce stable crystals or scale to "sheets".
P(generalises) = 0.8
Big | True = 4.5
Big * P(generalises) = 3.6
Physics
Engineering
Nanotechnology
Good
P(generalises) × big | true
0.8 × 4.5 = 3.6

The first observation of the "hexatic" phase of matter (a 2D metal between solid and liquid)

+
In an atomically-thin crystal of silver iodide. Confirms the theory that won the 2016 Nobel.
P(generalises) = 0.8
Big | True = 4.5
Big * P(generalises) = 3.6
Physics
Statistical mechanics
Good
P(generalises) × big | true
0.6 × 6 = 3.6

Chemical synthesis of artificial starch from carbon dioxide without cells

+
Still used engineered proteins. Seems to be similar to their 2021 process but cleaner.
P(generalises) = 0.6
Big | True = 6
Big * P(generalises) = 3.6
Chemistry
Good
P(generalises) × big | true
0.6 × 6 = 3.6

AI designs antibodies that can turn on or off membrane signaling proteins implicated in many diseases

+
G protein-coupled receptors are associated with retinitis, diabetes, hypertension, heart failure, obesity, and Alzheimer's. Nabla Bio uses test-time compute scaling to go through multiple rounds of generation and filtering of viable antibody structures (VHH, small single-chain antibodies). In contrast, another protein design company, Chai, relies on a strong base model that can zero-shot full-length drug-like IgG antibody design.
P(generalises) = 0.6
Big | True = 6
Big * P(generalises) = 3.6
Biology
AI
Protein design
Mixed
P(generalises) × big | true
0.9 × 4 = 3.6

The first "x-ray transient" (a star with periodic emissions in both radio and x-ray wavelengths)

+
ASKAP J1832-0911 is a long-period transient: it emits pulses of radio waves and X-rays every 44 minutes. It's around 15,000 light-years from us.
P(generalises) = 0.9
Big | True = 4
Big * P(generalises) = 3.6
Physics
Astronomy
Good
P(generalises) × big | true
0.9 × 4 = 3.6

The most energetic neutrino ever observed just smashed into the Mediterranean

+
i.e. the biggest in 55 years of looking. It left a signal picked up by the K3Mnet neutrino telescope 2400m under the sea
P(generalises) = 0.9
Big | True = 4
Big * P(generalises) = 3.6
Physics
Good
P(generalises) × big | true
0.8 × 4.5 = 3.6

Transcription factors bind to overlapping networks of weaker sites instead of single high-affinity DNA sites

+
A new model of transcription factor activity: TF binding depends on the sum of contributions from many overlapping sites
P(generalises) = 0.8
Big | True = 4.5
Big * P(generalises) = 3.6
Biology
Genetics
Transcription factors
Good
P(generalises) × big | true
0.6 × 6 = 3.6

AI for antibiotic design

+
7 of 24 AI-designed and custom-synthesized compounds show selective antibacterial activity, including against Neisseria gonorrhoeae and Staphylococcus aureus.
P(generalises) = 0.6
Big | True = 6
Big * P(generalises) = 3.6
Biology
AI
Small molecule design
Mixed
P(generalises) × big | true
0.7 × 5 = 3.5

AI proteins with massively increased mechanical strength; others survive 150°C

+
The resulting SuperMyo proteins have 4 times more resistance to unfolding forces than the source. Note that these properties were in two different synthetic proteins and that (a tiny number of) natural proteins have similar properties. F553's strength, seems to come at the expense of refolding ability. The RFdiffusion model generates the backbone, ProteinMPNN designs the sequence, ESM and AlphaFold predict its structure, and steered molecular dynamics do the in-silico screening. Result is systematic lengthening of β-sheet peptides and maximising the hydrogen bonds between them.
P(generalises) = 0.7
Big | True = 5
Big * P(generalises) = 3.5
AI
Biology
Protein design
Good
P(generalises) × big | true
0.7 × 5 = 3.5

First patients treated with safer proton arc radiotherapy

+
n=10 head and neck cancer patients. Proton beams already do much less damage to healthy tissue behind the target tumour; proton arcs also do much less damage to the tissue in front. They use about 20% less total radiation. Also much faster, since you can cover a volume with one session rather than triangulating with several beams.
P(generalises) = 0.7
Big | True = 5
Big * P(generalises) = 3.5
Physics
Medicine
Engineering
Mixed
P(generalises) × big | true
0.7 × 5 = 3.5

The CDC is likely to stop all testing on monkeys next year

+
P(generalises) = 0.7
Big | True = 5
Big * P(generalises) = 3.5
Biology
Medicine
Good
P(generalises) × big | true
0.7 × 5 = 3.5

Human chess grandmasters learned new useful concepts extracted automatically from AlphaZero

+
n=4 chess grandmasters
P(generalises) = 0.7
Big | True = 5
Big * P(generalises) = 3.5
AI
Human-AI
Good
P(generalises) × big | true
0.7 × 5 = 3.5

The "NanoKnife" electrocutes prostate tumours instead of cutting them out

+
Traditional surgery leads to incontinence in 36% of men with prostate cancer and erectile dysfunction in 95%. the PRESERVE Trial (n=121) . Doesn't seem to have had a control group.
P(generalises) = 0.7
Big | True = 5
Big * P(generalises) = 3.5
Biology
Medicine
Oncology
Good
P(generalises) × big | true
0.7 × 5 = 3.5

The first antiviral against dengue fever, about 60% effective

+
14 million dengue cases a year, 10,000 deaths. Phase 2a human challenge trial (n=21). They tested prophylaxis but it might work as treatment too.
P(generalises) = 0.7
Big | True = 5
Big * P(generalises) = 3.5
Biology
Medicine
Tropical medicine
Good
P(generalises) × big | true
0.7 × 5 = 3.5

First restoration of sight after macular degeneration using implants and special glasses

+
Partial restoration of sight after macular degeneration. Eye implants and some special glasses, clinical trial with nonrandom n=32, 1 year followup.
P(generalises) = 0.7
Big | True = 5
Big * P(generalises) = 3.5
Biology
Medicine
Vision
Good
P(generalises) × big | true
0.7 × 5 = 3.5

Insulin delivery through a topical cream

+
Successful non-invasive insulin delivery. Experiments in mice with Type I diabetes and in streptozotocin-induced diabetic minipigs. Topically administered polymer-conjugated insulin lowered blood glucose within 1-2 hours and maintained normal glucose levels for ~12 hours.
P(generalises) = 0.7
Big | True = 5
Big * P(generalises) = 3.5
Biology
Medicine
Diabetes
Good
P(generalises) × big | true
0.7 × 5 = 3.5

A battery-free intracellular delivery patch for internal organs

+
NanoFLUID is a chipless patch that can be sutured to the target organ (liver, mammary gland in mice) and contains the following components: a nanopore, a microchannel and a microelectrode. It can electroporate target cells and deliver drug payloads intracellularly (small molecules or gene therapy). No human studies yet.
P(generalises) = 0.7
Big | True = 5
Big * P(generalises) = 3.5
Biology
Medicine
Gene therapy, drug delivery
Good
P(generalises) × big | true
0.7 × 5 = 3.5

We now know when aerobic bacteria arose (3.2 billion years ago, 900 million years before photosynthesis was widespread)

+
Molecular clocks suggest that aerobic metabolism first emerged 900 million years before the Great Oxidation Event. "aerobic respiration evolved before widespread atmospheric oxygenation and may [therefore] have facilitated the evolution of oxygenic photosynthesis." Clock calibration uncertainties and model assumptions (especially horizontal gene transfer rates) still limit our confidence.
P(generalises) = 0.7
Big | True = 5
Big * P(generalises) = 3.5
Biology
Evolution
Aerobic bacteria
Good
P(generalises) × big | true
0.2 × 5 = 3.5

Reversal of "ten years" of (proxies for) aging in one month by manipulating just one gene

+
Preprint shows epigenetic rejuvenation in primary cells by manipulating the expression of just one gene. “Ten years” of reversed ageing in one month of overexpression. This is as measured by “epigenetic clocks” and gene expression profiles, a proxy for actual aging. Shown in dermal and lung fibroblasts and keratinocytes.
P(generalises) = 0.2
Big | True = 5
Big * P(generalises) = 3.5
Biology
Medicine
Aging
Good
P(generalises) × big | true
0.7 × 5 = 3.5

One man bitten hundreds of times enables the creation of extremely effective antivenoms

+
For 18 years, Tim Friede deliberately injected himself with snake venom and took hundreds of bites from highly lethal snakes. His blood now contains antibodies to these venoms, making him hyperimmune to them. Researchers used his blood samples to make an antibody library and select for antibodies with the most potent venom-neutralizing activity. They identified 2 antibodies which in combination with a previously used small molecule proved highly effective against 13 out of 19 elapid venoms (and partially effective against all 19).
P(generalises) = 0.7
Big | True = 5
Big * P(generalises) = 3.5
Biology
Medicine
Antibody engineering
Good
P(generalises) × big | true
0.7 × 5 = 3.5

A natural experiment provides evidence that the shingles vaccine protects against dementia. 3.5% absolute risk reduction

+
7 year follow-up on 79 year olds vs 80 year olds.
P(generalises) = 0.7
Big | True = 5
Big * P(generalises) = 3.5
Biology
Medicine
Vaccines / dementia
Good
P(generalises) × big | true
0.5 × 7 = 3.5

An IgE vaccine protects against dangerous allergies in mice

+
n=16. A Xolair drug bound to IgE fragments! This is already possible with expensive and repeated monoclonal antibody infusions.
P(generalises) = 0.5
Big | True = 7
Big * P(generalises) = 3.5
Biology
Medicine
Good
P(generalises) × big | true
0.5 × 7 = 3.5

A new class of flu antiviral outperforms a flu jab

+
Phase 2b trial (n=5000) of single-dose injection of CD388 found up to 76% protection against influenza.
P(generalises) = 0.5
Big | True = 7
Big * P(generalises) = 3.5
Biology
Medicine
Flu
Good
P(generalises) × big | true
0.5 × 7 = 3.5

Tiny demo of a 90% effective malaria vaccine which only takes one dose

+
However, the Leiden study was n=15. The followup PfSPZ-LARC2 study won't be finished until 2027 and is also n=22(!). Last year's R21 vaccine was 75% effective but takes 4 doses.
P(generalises) = 0.5
Big | True = 7
Big * P(generalises) = 3.5
Biology
Medicine
Vaccines
Good
P(generalises) × big | true
0.7 × 4.8 = 3.36

A protein language model generates new "superior" Yamanaka factors with a suggestion hit rate >30%, compared to <10% in normal screening

+
A protein language model generates new “superior” Yamanaka factors. Suggestion hit rate was >30%, compared to <10% in normal screening.
P(generalises) = 0.7
Big | True = 4.8
Big * P(generalises) = 3.36
AI
Biology
Protein design
Good
P(generalises) × big | true
0.65 × 5 = 3.25

First conversion of a human somatic cell into an egg cell which then produced an embryo

+
Low success rates and all embryos were aneuploid. The main problem before this work was that doing nuclear transfer (as many have before) leaves you with too many chromosomes. A few novelties involved: first “mitomeiosis” (electroporation + roscovitine). Their new method of artificial fertilisation gets close to natural baseline (78% egg activation rate). But 9% embryogenesis success rate and all embryos were aneuploid. Random chromosome segregation, no crossover recombination. Long way off from usable in vitro gametogenesis.
P(generalises) = 0.65
Big | True = 5
Big * P(generalises) = 3.25
Biology
Medicine
Reproductive medicine
Good
P(generalises) × big | true
0.4 × 8 = 3.2

In theory, you can detect quantum phenomena nondestructively by causing weird heat flows

+
Mathematical result: Entangle some quantum memory with the target system and a simple heat sink the system is not entangled with. If the target is quantum system, there will be more heat flow between the target and the sink than classical thermodynamics predicts. We don't measure the quantum system directly, so we don't cause decoherence. This would allow us to e.g. spot a fake quantum computer.
P(generalises) = 0.4
Big | True = 8
Big * P(generalises) = 3.2
Physics
Mathematics
Good
P(generalises) × big | true
0.8 × 4 = 3.2

CERN drove 105 trapped protons across campus in a truck

+
Seems to be the second particle relocation study on record, and the first on protons. The point is to be soon able to move antimatter for other labs to study,
P(generalises) = 0.8
Big | True = 4
Big * P(generalises) = 3.2
Physics
Engineering
Bad
P(generalises) × big | true
0.8 × 4 = 3.2

Retraction of a 2022 Science paper claiming "direct imaging of neuronal activity,” (DIANA), using fMRI to capture individual neuronal activity

+
In the original paper, a Korean team led by Jang-Yeon Park at Sungkyunkwan University, used MRI to capture a signal in the somatosensory cortex of mice 25 milliseconds after shocking their whisker pads. The results were found to be unreplicable. The authors reanaltyzed the data, admitted artifacts in their measurements and retracted the paper.
P(generalises) = 0.8
Big | True = 4
Big * P(generalises) = 3.2
Anti-breakthrough
Metascience
Neuroscience
Bad
P(generalises) × big | true
0.8 × 4 = 3.2

40% of animal studies of brain bleeding had problematic images

+
Out of 608 articles analyzed in a systematic review of subarachnoid hemorrhage, 243 were found to contain problematic images. Of those, 19 were retracted and 55 corrected.
P(generalises) = 0.8
Big | True = 4
Big * P(generalises) = 3.2
Anti-breakthrough
Research fraud
Good
P(generalises) × big | true
0.8 × 4 = 3.2

The oldest protein ever sequenced was previously 4 million years old, but now a 20 million year old protein has been sequenced

+
Previously the oldest protein ever sequenced was ~3.7 million years old. Now it's 21-24 mya. The protein sequences in question are from the enamel of a rhinocerotid from Canada’s High Arctic and can help deduce phylogeny of these animals.
P(generalises) = 0.8
Big | True = 4
Big * P(generalises) = 3.2
Biology
Genetics
Paleoproteomics
Good
P(generalises) × big | true
0.8 × 4 = 3.2

You can just drain a haematoma through a needle rather than do brain surgery

+
n=51 patients with an average 98-day followup.
P(generalises) = 0.8
Big | True = 4
Big * P(generalises) = 3.2
Biology
Medicine
Hematomas
Good
P(generalises) × big | true
0.8 × 4 = 3.2

A one-time gene therapy curing haemophilia B has been approved in the US

+
The BEQVEZ Phase III trial was n=45 with <=3 years of follow-up
P(generalises) = 0.8
Big | True = 4
Big * P(generalises) = 3.2
Biology
Medicine
Genetics
Good
P(generalises) × big | true
0.8 × 4 = 3.2

A new quantum algorithm computes topological knot invariants and could one day be used for DNA, proteins, polymers, chemistry, and quantum computing

+
New quantum algorithm, for computing topological knot invariants. Not general, since it requires spectral gap assumptions. Complexity is uncertain; they give preliminary numerical evidence for it being polynomial-time. Could one day be used for DNA, proteins, polymer, cool chemistry, and (ha) quantum computing. There are now 75 known quantum algorithms.
P(generalises) = 0.8
Big | True = 4
Big * P(generalises) = 3.2
Mathematics
Good
P(generalises) × big | true
0.8 × 4 = 3.2

The first systematic discovery of new families of unstable singularities includes multiple new unstable self-similar solutions for nonlinear partial differential equations

+
first systematic discovery of new families of unstable singularities… multiple new, unstable self-similar solutions for the incompressible porous media equation and the 3D Euler equation with boundary… simple empirical asymptotic formula relating the blow-up rate to the order of instability… near double-float machine precision, attaining a level of accuracy constrained only by the round-off errors of the GPU hardware. This level of precision meets the requirements for rigorous mathematical validation via computer-assisted proofs. nonlinear partial differential equations (PDEs)
P(generalises) = 0.8
Big | True = 4
Big * P(generalises) = 3.2
Mathematics
Good
P(generalises) × big | true
0.5 × 6 = 3

Slapping a phage patch on your food might reduce food poisoning risk

+
P(generalises) = 0.5
Big | True = 6
Big * P(generalises) = 3
Biology
Medicine
Virology
Good
P(generalises) × big | true
0.5 × 6 = 3

Head injuries induce genetic damage similar to Alzheimer's

+
P(generalises) = 0.5
Big | True = 6
Big * P(generalises) = 3
Biology
Medicine
Genetics
Good
P(generalises) × big | true
0.6 × 5 = 3

Tiny robot steered through the arteries using magnets to deliver medicine to highly specific regions

+
Magnetic microrobot
P(generalises) = 0.6
Big | True = 5
Big * P(generalises) = 3
Biology
Medicine
Engineering
Good
P(generalises) × big | true
0.6 × 5 = 3

Rising Internet access reduces prevalence of female genital mutilation

+
P(generalises) = 0.6
Big | True = 5
Big * P(generalises) = 3
Social
Good
P(generalises) × big | true
0.5 × 6 = 3

Two promising drugs to prevent secondary and post-surgical stroke

+
12 million strokes a year. The target, clotting factor XI, is a big deal because it promises to reduce dangerous clots without increasing the risk of internal bleeding. No actual results published for asundexian yet.
P(generalises) = 0.5
Big | True = 6
Big * P(generalises) = 3
Biology
Medicine
Good
P(generalises) × big | true
0.5 × 6 = 3

Challenge trial on a salmonella vaccine showed roughly 70% effectiveness

+
Challenge trial (n=72) on a salmonella vaccine showed roughly 70% effectiveness. There are currently no effective vaccines for it. 2 million fever cases a year, 150k deaths.
P(generalises) = 0.5
Big | True = 6
Big * P(generalises) = 3
Biology
Medicine
Vaccines
Good
P(generalises) × big | true
0.5 × 6 = 3

Biggest ever run of de novo antibody design and in silico screening produces 55 new binding antibodies

+
A specialised version of AlphaFold2, mBER. 1.15 million candidate antibodies. 145 targets, 45% of which got a “design success” binder, and in vitro a 38% binding rate. Open sourced.
P(generalises) = 0.5
Big | True = 6
Big * P(generalises) = 3
AI
Medicine
Antibody design
Good
P(generalises) × big | true
0.5 × 6 = 3

A candidate "gene drive" for eliminating malaria reduced parasite hosting from 80% to 30%

+
A minimal change to the mosquito genome, a single-amino-acid, appears to make the mosquito refractory to the common malaria parasite without affecting the mosquito's fitness. If being infected has any cost for the mosquito (e.g. mere weight), then this edit would spread automatically and prevent malaria transmission.
P(generalises) = 0.5
Big | True = 6
Big * P(generalises) = 3
Biology
Medicine
Genetic engineering
Good
P(generalises) × big | true
0.99 × 3 = 2.97

Philosopher's stone: The LHC currently produces 89,000 gold nuclei per second from lead-lead collisions

+
Gold nuclei emerge from the collision with very high energy and hit the LHC beam pipe or collimators at various points downstream, where they immediately fragment into single protons, neutrons and other particles. The gold exists for just a tiny fraction of a second though.
P(generalises) = 0.99
Big | True = 3
Big * P(generalises) = 2.97
Physics
Good
P(generalises) × big | true
0.98 × 3 = 2.94

Sets containing line segments in all directions must have full dimension in 3D

+
May not extend easily to higher dimensions. Connections to harmonic analysis, restriction theory.
P(generalises) = 0.98
Big | True = 3
Big * P(generalises) = 2.94
Mathematics
Bad
P(generalises) × big | true
0.95 × 3 = 2.85

De facto retraction of a 2024 economics paper claiming many things about AI speedups in materials science

+
We bought this one last year! Pardon.
P(generalises) = 0.95
Big | True = 3
Big * P(generalises) = 2.85
Anti-breakthrough
Metascience
Economics
Good
P(generalises) × big | true
0.4 × 7 = 2.8

A tiny number of people are functionally cured of HIV. The antibodies responsible may have been identified

+
Technically, "sustained control" of the virus. Single-arm n=10 trial. Passive transfer of broadly neutralizing antibodies appears to help CD8+ T cells totally control the level of virus, though not to eliminate it.
P(generalises) = 0.4
Big | True = 7
Big * P(generalises) = 2.8
Biology
Medicine
Immunology
Good
P(generalises) × big | true
0.4 × 7 = 2.8

An E. coli vaccine is currently undergoing Phase III human testing

+
E. coli is the second-most lethal bacterium in the world, with about a million deaths a year. There are currently no effective vaccines for it. About 10% of food poisoning cases are in the crosshairs.
P(generalises) = 0.4
Big | True = 7
Big * P(generalises) = 2.8
Biology
Medicine
Vaccines
Good
P(generalises) × big | true
0.4 × 7 = 2.8

“Bridge editing” is a possible replacement for CRISPR with less mess and damage

+
“Bridge editing” is a possible replacement for CRISPR with less mess and damage. “write massive changes into the right place in the human genome”. There’s a naturally occurring recombinase enzyme that is more programmable than CRISPR because it doesn’t rely on noisy cell repair mechanisms coming in to actually do the intended edit. This paper found one that works ok in human cells and then engineered it intensely and then tried excising a disease gene. Insertions, excisions, and inversions at scale. It’s still only 82% specificity, but CRISPR was at 30% at a similar point in its development.
P(generalises) = 0.4
Big | True = 7
Big * P(generalises) = 2.8
Biology
Medicine
Gene editing
Good
P(generalises) × big | true
0.7 × 4 = 2.8

New AI method for cryogenic-electron microscopy

+
End-to-end and efficient CryoFold (E3-CryoFold) is a deep learning method that combines 3D and sequence transformers. Empirically E3-CryoFold shows 4x improvement in the modeling score compared to earlier models.
P(generalises) = 0.7
Big | True = 4
Big * P(generalises) = 2.8
AI
Structural biology
Cryo-EM
Good
P(generalises) × big | true
0.7 × 4 = 2.8

Life glows: DNA emits weak photons under specific conditions

+
Tested in purified barley DNA, under optimized temperature and pH conditions
P(generalises) = 0.7
Big | True = 4
Big * P(generalises) = 2.8
Biology
Bioelectricity
Good
P(generalises) × big | true
0.7 × 4 = 2.8

Scientists discover a new colour ("olo") by firing lasers into their retinas

+
Proof-of-principle showing that it is possible to directly control the human eye’s photoreceptor activity via cell-by-cell light delivery, selectively activating M cones only.
P(generalises) = 0.7
Big | True = 4
Big * P(generalises) = 2.8
Physics
Good
P(generalises) × big | true
0.5 × 5.5 = 2.75

Mutating the world's most common enzyme, RuBisCO, boosted plant growth by 50% when injected

+
The most common enzyme on Earth is RuBisCO, crucial to photosynthesis and therefore to most plant growth. It’s weirdly inefficient. An Australian team used a bacterium to screen Rubisco mutations, finding two big ones. Genetic engineering failed, but just injecting the enzyme A242V boosted tobacco growth by 50%. See also the MIT directed evolution study.
P(generalises) = 0.5
Big | True = 5.5
Big * P(generalises) = 2.75
Biology
Agriculture
Genetic engineering
Good
P(generalises) × big | true
0.9 × 3 = 2.7

The most realistic molecular picture of supercooled water to date

+
P(generalises) = 0.9
Big | True = 3
Big * P(generalises) = 2.7
Physics
Good
P(generalises) × big | true
0.9 × 3 = 2.7

Tylos has the most extreme planetary climate yet discovered, including the fastest winds (Mach 27)

+
P(generalises) = 0.9
Big | True = 3
Big * P(generalises) = 2.7
Physics
Astronomy
Good
P(generalises) × big | true
0.9 × 3 = 2.7

Detection of hydrocarbons on Enceladus, possibly oxides and nitrides

+
Detection of hydrocarbons on Enceladus, being stretched very thin to be a “prospect for life” on the moon. Maybe also oxides and nitrides. You people really really wish for terrible news.
P(generalises) = 0.9
Big | True = 3
Big * P(generalises) = 2.7
Physics
Astronomy
Good
P(generalises) × big | true
0.6 × 4.5 = 2.7

A bacterium with a possible causal role in major depressive disorder

+
The orignial connection was reported in a 2022 study. This year's publication pointed to a possible mechanism: Morganella morganii produces metabolites that induce pro-inflammatory immune response. These metabolites are unusual phospholipids, similar to endogenous cardiolipins. It was found that an environmental contaminant, diethanolamine (DEA), can substitute for a native sugar alcohol in these lipids. Chronic inflammation and increased IL-6 levels are associated with MDD, and the bacterial metabolites could act through this pathway.
P(generalises) = 0.6
Big | True = 4.5
Big * P(generalises) = 2.7
Biology
Medicine
Major depressive disorder
Good
P(generalises) × big | true
0.5 × 5 = 2.5

A functioning qubit made of protein

+
An optically addressable spin qubit in yellow fluorescent protein. They also put the protein into mammalian cells, an incredibly structurally complicated environment. They also put into a bacterium and got it somewhat working at room temperature. The point is to have programmable biosensors.
P(generalises) = 0.5
Big | True = 5
Big * P(generalises) = 2.5
Biology
Quantum computing
Good
P(generalises) × big | true
0.5 × 5 = 2.5

Claim of the first cellular microbe with no energy metabolism

+
The most dependent cell. Preprint finding of an archaea, Sukunaarchaeum mirabile, with the most extreme metabolic reduction known in a cellular lifeform - only informational metabolism (i.e. replication) remains. This is closer to a virus than any previously known life.
P(generalises) = 0.5
Big | True = 5
Big * P(generalises) = 2.5
Biology
Microbiology
Anomalies
Good
P(generalises) × big | true
0.5 × 5 = 2.5

Neutral-atom quantum computers start to take over

+
A 448-qubit machine doing computation (on at least 256 qubits); 6,100 atoms in an array and coherent, though no computation done on them.
P(generalises) = 0.5
Big | True = 5
Big * P(generalises) = 2.5
Physics
Engineering
Computing
Good
P(generalises) × big | true
0.5 × 5 = 2.5

AI design of high-frequency electronic circuits faster and more efficiently than traditional methods

+
CNN. "universal inverse design approach for arbitrary-shaped complex multi-port electromagnetic structures with designer radiative and scattering properties, co-designed with active circuits" "CNN based forward model for prediction of scattering parameters of arbitrary shaped passive EM structures, and an approach to inverse synthesis utilizing this forward model"
P(generalises) = 0.5
Big | True = 5
Big * P(generalises) = 2.5
AI
Computing
Inverse design
Good
P(generalises) × big | true
0.5 × 5 = 2.5

Various claims of automated AI scientists

+
It's mostly really good literature review for now
P(generalises) = 0.5
Big | True = 5
Big * P(generalises) = 2.5
AI
Good
P(generalises) × big | true
0.5 × 5 = 2.5

New non-hormonal hair loss treatment might work in a month instead of a year

+
Phase 2a trial (n=45) of PP405 in alopecia patients. "At week eight, only four weeks after the completion of treatment in men with a higher degree of hair loss, 31% of those treated with PP405 exhibited a greater than 20% increase in hair density, compared to 0% of patients responding in the placebo group. Typically, visible hair regrowth requires 6-12 months of continuous therapy, suggesting PP405 may yield a more rapid clinical response than current treatments. "
P(generalises) = 0.5
Big | True = 5
Big * P(generalises) = 2.5
Biology
Medicine
Good
P(generalises) × big | true
0.5 × 5 = 2.5

Skin cells turned into egg cells, possibly enabling gay couples to have biological children

+
Somatic cell nuclear transfer can convert somatic (non-germ) cells into oocytes but with a diploid chromosome count. Still issues with cell division (random chromosomal segregation & no crossover). Embryos formed after fertilization of the SCNT oocytes progressed through early stages of cel division.
P(generalises) = 0.5
Big | True = 5
Big * P(generalises) = 2.5
Biology
Medicine
Somatic oogenesis
Good
P(generalises) × big | true
0.5 × 5 = 2.5

A nematode survived being frozen in permafrost for 46,000 years and then resumed reproduction

+
If we are able to figure out how it does this, it could have applications for organ preservation and food preservation.
P(generalises) = 0.5
Big | True = 5
Big * P(generalises) = 2.5
Biology
Medicine
Cryonics
Good
P(generalises) × big | true
0.5 × 5 = 2.5

New type of microscope using a normal continuous wave laser (40x cheaper) instead of expensive ultrafast femtosecond-pulse laser systems

+
40x cheaper machine, maybe 100x cheaper beam. Tested for now with one particular fluorescent protein. “optical sectioning” - seeing inside tissue noninvasively. Preprint “Shooting lasers at mice brains”. “2 photon microscope, using dramatically cheaper light sources. In the limit this tech could image single neurons at the scale of the whole cortex and every synapse of an individual neuron.”
P(generalises) = 0.5
Big | True = 5
Big * P(generalises) = 2.5
Hardware
Neuroscience
Good
P(generalises) × big | true
0.5 × 5 = 2.5

Demo of non-volatile memory with access speeds even faster than the fastest volatile memory

+
P(generalises) = 0.5
Big | True = 5
Big * P(generalises) = 2.5
Hardware
Computing
Good
P(generalises) × big | true
0.5 × 5 = 2.5

Electric vehicles now last as long as petrol ones

+
Regression on UK MOT test data (n=29m vehicles). The headline result is descriptive though.
P(generalises) = 0.5
Big | True = 5
Big * P(generalises) = 2.5
Hardware
Good
P(generalises) × big | true
0.98 × 2.5 = 2.45

The first known polyhedron that can't tunnel through itself

+
It thus does not have Rupert's property, so it's a Noperthedron
P(generalises) = 0.98
Big | True = 2.5
Big * P(generalises) = 2.45
Mathematics
Good
P(generalises) × big | true
0.6 × 4 = 2.4

Bees "empathise" with each other

+
First demonstration of affective contagion in any invertebrate.
P(generalises) = 0.6
Big | True = 4
Big * P(generalises) = 2.4
Biology
Zoology
Delight
Bad
P(generalises) × big | true
0.8 × 3 = 2.4

The original evidence for cognitive dissonance theory is not evidence for it

+
The case study underlying Festinger's classic cognitive dissonance theory does not support the theory
P(generalises) = 0.8
Big | True = 3
Big * P(generalises) = 2.4
Anti-breakthrough
Metascience
Psychology
Good
P(generalises) × big | true
0.8 × 3 = 2.4

Wild chimps ingest alcohol equivalent to a pint of lager a day (via fermented fruit)

+
Supports the “drunken monkey” hypothesis: human attraction to alcohol possibly comes from an ancestral association of ethanol with ripe fruits
P(generalises) = 0.8
Big | True = 3
Big * P(generalises) = 2.4
Biology
Zoology
Delight
Good
P(generalises) × big | true
0.4 × 6 = 2.4

AI generator for antibodies against specific protein targets

+
If you test 70 designs per antigen you succeed around 10% of the time (pass@70). Notably this was all inference compute (300 H100-hours per run) on AbMPNN and AlphaFold3, no new model training. (AbMPNN is already finetuned for antibody prediction.)
P(generalises) = 0.4
Big | True = 6
Big * P(generalises) = 2.4
AI
Medicine
Antibody design
Good
P(generalises) × big | true
0.6 × 4 = 2.4

The first lab-grown salmon is now available. It's about $12 extra

+
Seems to be just in a couple of restaurants for now. Price could be heavily subsidised.
P(generalises) = 0.6
Big | True = 4
Big * P(generalises) = 2.4
Biology
Animal welfare
Lab grown meat
Good
P(generalises) × big | true
0.6 × 4 = 2.4

An AI simulator for thousands of molecules that runs well enough on a Macbook

+
Missed last year: Orb is an AI simulator for chemical potentials allowing for thousands of simultaneous molecules. Runs well enough on a Macbook.
P(generalises) = 0.6
Big | True = 4
Big * P(generalises) = 2.4
Chemistry
AI
Good
P(generalises) × big | true
0.3 × 8 = 2.4

Ideas might not be getting harder to find

+
P(generalises) = 0.3
Big | True = 8
Big * P(generalises) = 2.4
Social
Good
P(generalises) × big | true
0.9 × 2.5 = 2.25

The unknotting number is not additive under connected sum

+
The first counterexamples to a conjecture from 1997.
P(generalises) = 0.9
Big | True = 2.5
Big * P(generalises) = 2.25
Mathematics
Good
P(generalises) × big | true
0.9 × 2.5 = 2.25

A new theory of summation

+
Uses the hyperreals to assign fine-grained values to divergent sums and integrals. Developed for ethical applications.
P(generalises) = 0.9
Big | True = 2.5
Big * P(generalises) = 2.25
Mathematics
Analysis
Summability
Good
P(generalises) × big | true
0.9 × 2.5 = 2.25

29th moon of Uranus discovered. Just 6 miles wide

+
Discovery of a new moon of Uranus with the James Webb Space Telescope. The 29th. It’s just 6 miles wide.
P(generalises) = 0.9
Big | True = 2.5
Big * P(generalises) = 2.25
Physics
Astronomy
Good
P(generalises) × big | true
0.9 × 2.5 = 2.25

This is what the aurora on Neptune looks like

+
P(generalises) = 0.9
Big | True = 2.5
Big * P(generalises) = 2.25
Physics
Astronomy
Delight
Good
P(generalises) × big | true
0.9 × 2.5 = 2.25

An old and important conjecture in harmonic analysis was falsified by a teenager

+
The Mizohata-Takeuchi conjecture is false
P(generalises) = 0.9
Big | True = 2.5
Big * P(generalises) = 2.25
Mathematics
Good
P(generalises) × big | true
0.5 × 4.5 = 2.25

AI-designed antivenoms for a kind of snakebite that currently only has expensive impractical treatment

+
"Three-finger toxins". Actually a pipeline of open models: uses a Baker Lab model, RFDiffusion, to design the protein backbone, then ProteinMPNN to design the sequence, and then screens the designs in silico using AlphaFold2 and Rosettafold. Only n=15 mice
P(generalises) = 0.5
Big | True = 4.5
Big * P(generalises) = 2.25
AI
Medicine
Protein design
Good
P(generalises) × big | true
0.3 × 7 = 2.1

First time a brain region was turned back on after "vitrification" (extremely rapid cooling)

+
"near-physiological" functional recovery of hippocampus activity. Preprint. n=12.
P(generalises) = 0.3
Big | True = 7
Big * P(generalises) = 2.1
Biology
Neuroscience
Cryonics
Mixed
P(generalises) × big | true
0.3 × 7 = 2.1

Generative biology on the way. 285 new phage viruses from a DNA language model

+
They generated 285 new phage virus genomes with a DNA language model. 16 of them successfully attacked the target E. coli strain and (crucially) not other strains. “The viable generated phages harbored hundreds of novel mutations, many of which do not map to any sequence seen in nature.” The old interpolation problem crops up here too though: the AI-generated phage genomes were all >90% similar to existing phages. Phages are very safe and beneficial, but the same methods apply to undesirable viruses.
P(generalises) = 0.3
Big | True = 7
Big * P(generalises) = 2.1
Biology
Medicine
Genome design
Good
P(generalises) × big | true
0.7 × 3 = 2.1

Bonobos have compositional language

+
This was previously one of many false claims for human exceptionalism. Specifically, the interest here is in nontrivial compositionality where one element of a combination can modify the other(s), rather than trivial compositionality (the meaning of a combination is a straghtforward sum of its parts). Bonobo vocalizations were found for the first time to exhibit non-trivial compositionality.
P(generalises) = 0.7
Big | True = 3
Big * P(generalises) = 2.1
Biology
Zoology
Animal communication
Good
P(generalises) × big | true
0.7 × 3 = 2.1

Spiral galaxies that rotate in the opposite direction to our galaxy are 50% more common than galaxies that rotate in the same direction

+
P(generalises) = 0.7
Big | True = 3
Big * P(generalises) = 2.1
Physics
Astronomy
Good
P(generalises) × big | true
0.8 × 2.5 = 2

Genetic design of new flowers is now a viable hobby

+
Engineering flowers of Arabidopsis thaliana, the main model studied in plant labs, using a combination of mutant lines and gene editing. Two conceptual steps of the process include genetic sculpting, where one envisions the future appearance of a flower, and genetic painting, where one finds specific mutations and gene engineering approaches from the existing research literature to attain the envisioned flower morphology and coloration.
P(generalises) = 0.8
Big | True = 2.5
Big * P(generalises) = 2
Biology
Genetics
Delight
Good
P(generalises) × big | true
0.8 × 2.5 = 2

First test of the "second law of black hole dynamics" using gravitational waves to see if total surface area decreased in a giant black hole merger (as expected, it didn't)

+
P(generalises) = 0.8
Big | True = 2.5
Big * P(generalises) = 2
Physics
Astronomy
Good
P(generalises) × big | true
0.5 × 4 = 2

The first brain-computer interface which generalises across people without tuning

+
P(generalises) = 0.5
Big | True = 4
Big * P(generalises) = 2
Hardware
Neuroscience
Good
P(generalises) × big | true
0.4 × 5 = 2

The philosopher's stone: a theoretically scaleable method for synthesizing gold

+
Lining a fusion reactor with lead which catches fast neutrons. At the start, before markets adjust, this could be worth about as much as the power output.
P(generalises) = 0.4
Big | True = 5
Big * P(generalises) = 2
Chemistry
Good
P(generalises) × big | true
0.4 × 5 = 2

First commercial company to breed tritium fuel

+
No details on how much
P(generalises) = 0.4
Big | True = 5
Big * P(generalises) = 2
Hardware
Energy
Good
P(generalises) × big | true
0.4 × 5 = 2

Tiny test (n=16) of a pancreatic cancer vaccine has good results

+
Recurrence dropping from 80% to 30%. 3 year followup.
P(generalises) = 0.4
Big | True = 5
Big * P(generalises) = 2
Biology
Medicine
Vaccines
Good
P(generalises) × big | true
0.75 × 2.5 = 1.9

Progress in the mirror symmetry program?: a criterion for the rationality of cubic fourfolds

+
Large preprint including, among other things, the claim that cubic fourfolds are irrational except when they are on certain Hassett divisors. (The rationality of cubic fourfolds in general is a major open problem since 1943.) Much novel machinery and other results inside.
P(generalises) = 0.75
Big | True = 2.5
Big * P(generalises) = 1.9
Mathematics
Algebraic geometry
Birational geometry
Good
P(generalises) × big | true
0.6 × 3 = 1.8

Infrared contact lenses allow people to see in the dark, including with their eyes closed

+
Infrared contact lenses allow people to see in the dark, including with their eyes closed.
P(generalises) = 0.6
Big | True = 3
Big * P(generalises) = 1.8
Hardware
Delight
Good
P(generalises) × big | true
0.3 × 6 = 1.8

A vaccine for fentanyl is beginning human clinical trials

+
n=40 ARMR Phase I/II trial in Leiden. 90%+ reduction in brain opioid response in mice. No CTIS entry yet. There are obvious risks in making yourself immune to strong painkillers.
P(generalises) = 0.3
Big | True = 6
Big * P(generalises) = 1.8
Biology
Medicine
Addiction
Good
P(generalises) × big | true
0.3 × 6 = 1.8

World's first 'thermodynamic computing chip' is ready for prototyping

+
P(generalises) = 0.3
Big | True = 6
Big * P(generalises) = 1.8
Physics
Computing
Good
P(generalises) × big | true
0.3 × 6 = 1.8

First successful transplant of a non-human lung into a human

+
Pig lung; recipient was alive but brain-dead. Given a shortage of donors, this could allow many more patients to receive transplants.
P(generalises) = 0.3
Big | True = 6
Big * P(generalises) = 1.8
Biology
Medicine
Organ transplants
Good
P(generalises) × big | true
0.3 × 6 = 1.8

Scalable evaporation based solar desalination

+
a grid of porous aerogels. "Unlike conventional evaporators that suffer over 40% reduction in evaporation performance with increasing size, the resulting aerogel... <5% reduction." Materials people make a lot of claims like this.
P(generalises) = 0.3
Big | True = 6
Big * P(generalises) = 1.8
Hardware
Good
P(generalises) × big | true
0.7 × 2.5 = 1.75

Some wolves are pollinators

+
Ethiopian wolves drink flower nectar, the first known instance in any large carnivore
P(generalises) = 0.7
Big | True = 2.5
Big * P(generalises) = 1.75
Biology
Zoology
Delight
Mixed
P(generalises) × big | true
0.7 × 2.5 = 1.75

Genetic explanation for how cuckoo eggs can mimic multiple host species' eggs

+
Substantial coevolution found, an arms race over egg mimicry
P(generalises) = 0.7
Big | True = 2.5
Big * P(generalises) = 1.75
Biology
Zoology
Bad
P(generalises) × big | true
0.7 × 2.5 = 1.75

The first European farmers disappeared in a continent-scale wave of violence around 5000BCE

+
n=34 headless corpses in one dig, more to come.
P(generalises) = 0.7
Big | True = 2.5
Big * P(generalises) = 1.75
Archaeology
Good
P(generalises) × big | true
0.7 × 2.5 = 1.75

First decipherment of a pre-Islam Eastern South Semitic script

+
(One of the two subtypes anyway)
P(generalises) = 0.7
Big | True = 2.5
Big * P(generalises) = 1.75
Archaeology
Linguistics
Good
P(generalises) × big | true
0.7 × 2.5 = 1.75

A smart pattern detector for enzyme cleavage sites in proteins

+
Peptide predictor is a program that identifies whether a secreted protein has prohormone activity based on the number of cleavage sites it contains. The cleavage sites are a combination of lysine and arginine residues.
P(generalises) = 0.7
Big | True = 2.5
Big * P(generalises) = 1.75
Biology
Biochemistry
Hormone processing
Bad
P(generalises) × big | true
0.5 × 3 = 1.5

In March, hundreds of US gov medical datasets were altered silently without a changelog

+
Mostly changing the word "gender" to the word "sex", but also changing "Social determinants” to “non-medical factors". Substantial changes in half of the 232 datasets studied. 13% of the 114 altered datasets included a log of the change.
P(generalises) = 0.5
Big | True = 3
Big * P(generalises) = 1.5
Social
Medicine
Good
P(generalises) × big | true
0.6 × 2.5 = 1.5

Grease rendering industry is forever: neanderthals were processing hundreds of big animals for bone marrow 125,000 years ago

+
Neanderthals in Germany were processing hundreds of big animals for bone marrow 125,000 years ago, a kind of industry. Likely boiled bones in organic containers. Difficult to distinguish from later processing.
P(generalises) = 0.6
Big | True = 2.5
Big * P(generalises) = 1.5
Archaeology
Good
P(generalises) × big | true
0.6 × 2.5 = 1.5

Ice can get slippery without melting, because of spontaneous mechanical stresses

+
Simulation suggests that ice is mostly slippery, not due to friction melting some water for you to slip on, but instead
P(generalises) = 0.6
Big | True = 2.5
Big * P(generalises) = 1.5
Physics
Good
P(generalises) × big | true
0.3 × 5 = 1.5

You might be able to get extra information from exotic "positronium" formed during a PET scan

+
P(generalises) = 0.3
Big | True = 5
Big * P(generalises) = 1.5
Hardware
Medicine
Good
P(generalises) × big | true
0.3 × 5 = 1.5

Galaxies may contribute measurable foreground contamination that challenges current cosmological interpretations

+
P(generalises) = 0.3
Big | True = 5
Big * P(generalises) = 1.5
Physics
Astronomy
Good
P(generalises) × big | true
0.2 × 7 = 1.4

Ozempic reverses epigenetic markers of ageing by 3 years

+
Phase 2 trial (n=84), 8 month followup.
P(generalises) = 0.2
Big | True = 7
Big * P(generalises) = 1.4
Biology
Medicine
Aging
Good
P(generalises) × big | true
0.5 × 2.5 = 1.25

Some spiders build scarecrow spiders against predators

+
Orb-weaving spiders create giant fake spiders in their own web, built from tiny bits of silk, animal carcasses, and plant debris,
P(generalises) = 0.5
Big | True = 2.5
Big * P(generalises) = 1.25
Biology
Zoology
Delight
Good
P(generalises) × big | true
0.5 × 2.5 = 1.25

Ancient DNA reveals a previously unknown branch of humanity, the Takarkori (modern Libya)

+
Sequencing of two mummies found in a cave. The lineage was still around in 5000 BCE and highly distinct from sub-Saharan Africans and Levantine farmers of the time.
P(generalises) = 0.5
Big | True = 2.5
Big * P(generalises) = 1.25
Biology
Archaeology
Human phylogeny
Good
P(generalises) × big | true
0.5 × 2.5 = 1.25

The quantum complexity class QMA is larger than QCMA

+
"it finally settles a problem that’s been open since 2002"
P(generalises) = 0.5
Big | True = 2.5
Big * P(generalises) = 1.25
Mathematics
Theoretical computer science
Quantum complexity theory
Good
P(generalises) × big | true
0.2 × 6 = 1.2

Some new drugs which treat long covid as an acquired autoimmune disorder

+
JAK (Janus kinase) inhibitors: abrocitinib (Cibinqo), baricitinib (Olumiant), and upadacitinib (Rinvoq). All three have previously approved for different conditions, so these are all cases of drug repurposing.
P(generalises) = 0.2
Big | True = 6
Big * P(generalises) = 1.2
Biology
Medicine
COVID
Good
P(generalises) × big | true
0.2 × 6 = 1.2

An attempt to separate the psychopharmaceutical effects of DMT from the hallucinogenic effect

+
Completed but unpublished Phase I trial (n=52) of a variant of 5-MeO-MiPT. the "least psychedelic psychedelic that's psychoactive." Grandiose claims: "If successful, psychiatry would gain for mental states what CRISPR gave genetics, what mRNA gave vaccines, and what CAR-T gave immunology: A programmable substrate”
P(generalises) = 0.2
Big | True = 6
Big * P(generalises) = 1.2
Biology
Medicine
Bad
P(generalises) × big | true
0.2 × 6 = 1.2

Proposal to nuke the ocean floor to pulverise basalt and accelerate carbon sequestration

+
P(generalises) = 0.2
Big | True = 6
Big * P(generalises) = 1.2
Environment
Geoengineering
Good
P(generalises) × big | true
0.2 × 6 = 1.2

First transplant cloned islet cells (the insulin-producing parts of the pancreas) into a patient and she's ok after a year

+
Nankai University team claim to have transplanted cloned islet cells (the insulin-producing parts of the pancreas) into a patient and she's ok after a year
P(generalises) = 0.2
Big | True = 6
Big * P(generalises) = 1.2
Biology
Medicine
Diabetes / organ transplants
Good
P(generalises) × big | true
0.2 × 5 = 1

The first clinical trial for a cure for any prion disease

+
The treatment is based on an siRNA (short interfering RNA, non-coding RNA that can degrade target RNAs via RNA interference). This would prevent prion proteins from being translated. For context, the first siRNA drug, patisiran for transthyretin amyloidosis already received FDA approval in 2018.
P(generalises) = 0.2
Big | True = 5
Big * P(generalises) = 1
Biology
Medicine
Prion diseases
Good
P(generalises) × big | true
0.2 × 5 = 1

We might be able to use particle accelerators to make nuclear waste into tritium fuel

+
"Computer simulations are performed to study the feasibility of firing proton beams onto spent fuel rods coated with molten lithium salt, which would initiate a fission process that converts lithium into tritium. In short, it is reigniting the fission process on the waste using an accelerator rather than in a reactor pressure vessel."
P(generalises) = 0.2
Big | True = 5
Big * P(generalises) = 1
Hardware
Energy
Good
P(generalises) × big | true
0.3 × 3 = 0.9

First recorded "xenoparity" in the wild, an animal giving birth to offspring of a different species

+
Began as parasitism and evolved into cross-species cloning for clean genome-level separation of fertility
P(generalises) = 0.3
Big | True = 3
Big * P(generalises) = 0.9
Biology
Zoology
Reproduction
Good
P(generalises) × big | true
0.1 × 8 = 0.8

An Ozempic-like reduces markers of aging in mice

+
n=18 middle-aged mice, half given the old-fashioned GLP-1 agonist exenatide for 30 weeks. "most effectively benefits physiological functions that exhibit more prominent age-related decline, including forelimb grip strength and rotarod performance". Effect comparable to rapamycin (i.e. some).
P(generalises) = 0.1
Big | True = 8
Big * P(generalises) = 0.8
Biology
Medicine
Longevity
Bad
P(generalises) × big | true
0.1 × 7 = 0.7

In areas with toxic levels of natural fluoride (mostly China), observational studies find that children have lower IQ scores. (NB: doesn't generalise!)

+
A meta-analysis of the fluoride-IQ link using mostly just China and India studies involving massive overdoses. For water fluoride <1.5 mg/L (WHO guideline), the association is null. Harm only appears when using urinary fluoride measures. But urinary fluoride conflates total exposure from all sources (water, food, toothpaste, tea, pollution from coal burning). You can't isolate the effect of water fluoridation. I² = 94% suggests they're pooling fundamentally different studies. As such, we think this doesn't generalise to developed world water systems.
P(generalises) = 0.1
Big | True = 7
Big * P(generalises) = 0.7
Biology
Public health
Toxicology
Good
P(generalises) × big | true
0.1 × 7 = 0.7

BDNF, a growth factor in the brain, may be a fast-acting depression treatment

+
Or rather the TrkB receptor which is involved in regulating BDNF. The study was done in rodents.
P(generalises) = 0.1
Big | True = 7
Big * P(generalises) = 0.7
Biology
Medicine
Neuroscience
Mixed
P(generalises) × big | true
0.1 × 7 = 0.7

Signs of low-temperature redox reactions in a Mars rock

+
More wishful Life on Mars talk. The low "P(generalises)" score here refers to using this as evidence for life
P(generalises) = 0.1
Big | True = 7
Big * P(generalises) = 0.7
Biology
Astronomy
Life on Mars
Good
P(generalises) × big | true
0.1 × 7 = 0.7

First "QI" stellarator design that could be used in a full-scale fusion power plant

+
first QI stellarator-based power plant design that simultaneously meets all physics and engineering constraints, as demonstrated through electromagnetic, structural, thermal, and neutronic simulations.
P(generalises) = 0.1
Big | True = 7
Big * P(generalises) = 0.7
Physics
Energy
Fusion
Good
P(generalises) × big | true
0.1 × 7 = 0.7

Progenitor cells counter aging in primates

+
Macaques given "senescence-resistant human mesenchymal progenitor cells" for around a year.
P(generalises) = 0.1
Big | True = 7
Big * P(generalises) = 0.7
Biology
Medicine
Aging
Good
P(generalises) × big | true
0.2 × 3 = 0.6

New gene therapy (AMT-130) with a putative 75% slowdown in Huntingdon's progression

+
Press release gives a new gene therapy AMT-130 a putative 75% slowdown in Huntingdon’s progression over 3 years. No proper results up yet
P(generalises) = 0.2
Big | True = 3
Big * P(generalises) = 0.6
Biology
Medicine
Huntington's
Good
P(generalises) × big | true
0.1 × 6 = 0.6

Attempt to derive dark energy from string theory

+
P(generalises) = 0.1
Big | True = 6
Big * P(generalises) = 0.6
Physics
Mixed
P(generalises) × big | true
0.2 × 2.5 = 0.5

The Native Americans may have originated in Japan, arriving in the Americas 20,000 years ago

+
P(generalises) = 0.2
Big | True = 2.5
Big * P(generalises) = 0.5
Archaeology
Good
P(generalises) × big | true
0.1 × 5 = 0.5

Tooth regrowing procedure enters human trials

+
Phase I trial of a monoclonal antibody for congenital anadontia (6 or more teeth missing or never growing). In mice, mutation or knockout of USAG-1 causes development of supernumerary teeth. Further experiments in mice showed that USAG-1 is a valid therapeutic target. Researchers at Kitano Hospital in Japan developed neutralizing antibodies against USAG-1 (TRG035). First-in-human studies are currently underway in healthy adult volunteers.
P(generalises) = 0.1
Big | True = 5
Big * P(generalises) = 0.5
Biology
Medicine
Dentistry
Good
P(generalises) × big | true
0.1 × 5 = 0.5

Circulatronics: a microscopic brain implant, a hybrid of immune cells and electronics, that you can inject

+
In mice
P(generalises) = 0.1
Big | True = 5
Big * P(generalises) = 0.5
Biology
Medicine
Neuromodulation
Good
P(generalises) × big | true
0.1 × 5 = 0.5

Claimed quantum advantage in a complicated physics task

+
Some of the peer reviewers were really not happy with the amount of exaggeration.
P(generalises) = 0.1
Big | True = 5
Big * P(generalises) = 0.5
Physics
Good
P(generalises) × big | true
0.1 × 4 = 0.4

Micronozzles could triple the proton energies of laser accelerators

+
Simulation shows unprecedentedly strong beams, 1 GeV
P(generalises) = 0.1
Big | True = 4
Big * P(generalises) = 0.4
Physics
Engineering
Good
P(generalises) × big | true
0.05 × 6 = 0.3

Dark photons: maybe you don't need wave-particle duality

+
P(generalises) = 0.05
Big | True = 6
Big * P(generalises) = 0.3
Physics
Mixed
P(generalises) × big | true
0.03 × 9 = 0.27

Very weak indicator of life spotted outside the solar system

+
"A possible biosignature gas" spotted on the exoplanet K2-18b. A 3-sigma result, so not formally claimed as a discovery. Dimethyl sulfide and dimethyl disulfide are mostly created through biotic processes on Earth but are easily created abiotically and are found throughout the galaxy. This seems desperate
P(generalises) = 0.03
Big | True = 9
Big * P(generalises) = 0.27
Biology
Astronomy
Extraterrestrial life
Good
P(generalises) × big | true
0.05 × 5 = 0.25

Claim that we have identified a bacterium that causes schizophrenia

+
Schizophrenia is heritable, but we have struggled to identify the genes involved, only explaining 32% of the variance with monstrously large polygenic scoring. Skolnick instead proposes that one unusual gut bacterium, Ruminococcus gnavus, is responsible through a specific amine pathway, effectively drugging you constantly over your life. MAYBE MAYBE this means it's curable. Obvious test is to see what happens (has already happened) in patients who have schizophrenia and also get full-spectrum antibiotics for other reasons.
P(generalises) = 0.05
Big | True = 5
Big * P(generalises) = 0.25
Biology
Medicine
Microbiology
Good
P(generalises) × big | true
0.05 × 5 = 0.25

Another unified field theory with some cool predictions

+
Mathematical result in service of the idea that electromagnetic radiation is really an oscillation of spacetime. In particular, the charge density is a field.
P(generalises) = 0.05
Big | True = 5
Big * P(generalises) = 0.25
Physics
Good
P(generalises) × big | true
0.01 × 2.5 = 0.025

Can I run DOOM on my soup?

+
Theoretical demonstration of the Turing completeness of certain exotic flows of fluid
P(generalises) = 0.01
Big | True = 2.5
Big * P(generalises) = 0.025
Physics
Mathematics
Delight