Racing the Machine to the Proof

On September 11, 2026, rumors spread that OpenAI had proved the unique games conjecture, one of the best-known open problems in computational complexity. MIT's Dor Minzer and his graduate students Yumou Fei and Shuo Wang were still writing up a related result. Three days later they posted a 95-page paper with a warning on the first page: it is "complete mathematically, but it is not in the shape we wished to share in." On October 6, OpenAI announced its proof along with 376 other results.

What the conjectures say

Both problems are about satisfying many constraints that pull against each other. In Subhash Khot's 2002 version, you color the nodes of a graph, and each edge has a rule: once one endpoint has a color, the other endpoint's color is fixed. The conjecture says that when the best coloring satisfies 99% of the rules, it can still be hard to find one that satisfies even a tiny fraction. Prasad Raghavendra showed in 2008 that if this is true, one classic algorithm is the best possible for every constraint problem with no perfect solution.

The conjecture has a gap, though. It says nothing about cases where all the constraints can be satisfied. To cover those, Khot proposed the 2-to-1 games variant, where each rule allows two colors on the other side. The claim is that finding even a weak approximation stays hard even when a perfect solution exists.

What the MIT team proved

Minzer, Fei and Wang proved a slightly weaker 4-to-1 version, which has almost the same consequences. Their strategy was to build a mathematical bridge from the coloring problem to one whose difficulty is already understood. Like Minzer's 2018 milestone, it goes through . Fei and Wang found a new way to construct those codes in 2025. Fitting it into the rest of the proof took repeated failures, until it worked in April 2026. "We took five failures and managed to staple this [new code] on top of them," Minzer said.

The list of high-confidence autism-linked genes has grown from 72 to 253, according to three companion preprints posted on bioRxiv in August and September. None has been peer-reviewed yet. The team, which includes Michael Talkowski of Harvard Medical School and Mark Daly of the Broad Institute, did the earlier 2022 study. That work searched rare coding variants in about 20,000 autistic people.

How the genes were found and sorted

The new analysis scanned the protein-coding DNA (exomes) of 62,429 autistic people from the Autism Sequencing Consortium and three other cohorts, including SPARK and the Simons Simplex Collection. It looked for de novo variants, which are new mutations present in a child but in neither parent.

The researchers then checked whether each gene is also linked to schizophrenia, bipolar disorder, epilepsy or other neurodevelopmental conditions. The genes fell into six clusters, which group into two broad camps:

• Autism-specific genes are enriched for synapse assembly and associated with immature interneurons.
• Genes shared with many conditions are enriched for transcriptional regulation, the control of which genes get switched on, and show stronger enrichment in excitatory neurons.

These camps track with outcomes. Children with a variant in a broadly shared gene were more likely to have developmental delays and to miss milestones. Daly cautions that the clusters are "probabilistic but not predictive": they cannot yet forecast an individual child's course.

Many molecules come in two mirror-image versions with the same atoms bonded in the same order, like left and right hands. This property is called chirality, and the two forms are labeled D and L. Most reactions make a 50-50 mix, but life uses only one form. Every organism's sugars are D, for example. Enzymes have binding sites shaped to fit one form, so they often can't work with its mirror image.

That leaves origin-of-life research with a puzzle. Either early life was less fussy and became selective slowly, or it arose where one form already dominated. The second option only moves the question back: how would chemistry produce that imbalance?

From theory to the bench

The physicist Frederick Charles Frank proposed several answers in a purely theoretical paper. A catalyst with a slight preference could build up a large excess over time. A chiral product could catalyze more of itself. Or it could suppress its mirror image. Timmer notes that the Nobel committee's writing suggests Frank might have shared the prize if he were still alive.

Henri Kagan (University of Paris-Sud) also started from theory, then found real reactions to test it. He worked with a catalyst that was itself chiral and found that in a mix of its D and L versions, one did most of the work. When that catalyst made a chiral product, the product came out heavily skewed toward one form. By the mid-1960s he had found three such reactions. Later researchers learned to switch chiral catalysts on or off by adding non-chiral chemicals, and that finer control made its way into pharmaceutical production.

In 1999, Richard Schulz and Scott Beach reported in JAMA that distressed caregivers were 63 percent more likely to die within four years than non-caregivers. The paper has been cited almost 5,300 times, nearly 300 of them in the past two years. A review published in September in Biopsychosocial Science and Medicine argues that other researchers have mostly failed to replicate it. The authors are UCLA neuroscientist Naomi Eisenberger and colleagues, and they say much of the research links caregiving with longer life instead.

The confound in the original study

The 1999 study followed more than 800 people. About half had disabled spouses, and some of those cared for their spouse while others did not. The other half had healthy spouses and served as controls. During the study, 103 participants died. Among caregivers who reported heavy strain, 17 percent died, compared with 9 percent of controls.

Eisenberger's team points to a result that got far less attention. People who had a disabled spouse but didn't care for them died at the same rate as the strained caregivers. In their reading, the risk comes from having a sick or disabled spouse, not from caregiving. Schulz himself coauthored a 2009 Psychological Science paper linking caregiving to lower mortality. It has fewer than 400 citations.

Stress without inflammation

Chronic stressors such as job burnout usually go along with systemic inflammation. Caregiving appears to be different. A 2020 study drew on a survey of more than 30,000 U.S. adults that included blood samples. Researchers tracked about 250 people as they became regular caregivers and measured six inflammation-related biomarkers. The markers did not rise, either compared with the same people before caregiving or compared with matched non-caregivers. That held even for the most stressed group, such as spouses caring for a partner with dementia.

In 2021, researchers reported that an optogenetic gene therapy helped one blind man see and count objects. A new study in the New England Journal of Medicine, published October 7, gave the same treatment to more people. It worked for some of them, though the vision they got back is very limited.

How the treatment works

All the participants had retinitis pigmentosa, an inherited disease that slowly kills the retina's light-sensing cells. The therapy doesn't try to repair those cells. It gives their job to other cells instead.

Botond Roska of the Institute of Molecular and Clinical Ophthalmology Basel and his colleagues injected genetic instructions for ChrimsonR, a light-sensitive protein that comes from algae, into each participant's worse-seeing eye. The instructions target retinal ganglion cells. These cells normally just pass visual signals along to the brain. Once they make ChrimsonR, they respond to amber light.

Ordinary light isn't enough to drive the treated cells, so participants wear custom goggles. A camera in the goggles picks up changes in brightness in front of the wearer. The goggles then project those changes onto the retina as pulses of amber light, which activate the treated cells. This is optogenetics: using genetic instructions to make cells respond to light. Pioneering work on the technique won the 2026 Nobel Prize in Physiology or Medicine on October 5.