Anthropic scientist made a new cell biology discovery by pointing a loop at raw microscopy and letting it delete 317 of its own findings by morning.
It was never asked to identify anything. It was asked what would have to be true, and what would prove it wrong.
> WATCH - raw microscopy in, no labels, no prior dataset, no hint of what to look for
> NOTICE - marks only the frames that break pattern, ignores everything that fits
> PROPOSE - every idea leaves with its assumption stated, never as a bare claim
> KILL - discards anything no bench experiment could ever disprove
> QUEUE - what survives becomes a protocol a human can actually run in the morning
320 hypotheses went in. 3 came out. The 3 were not the most convincing ones.
They were the only ones that could fail, which is the entire filter and the entire point.
Everything else read like a finding. Plausible is free, and a hundred plausible results by morning is not a result, it is noise dressed up as progress.
One of the 3 survived nine consecutive nights and is now the discovery - a division timing shift nobody had a reason to look for, confirmed on a bench by someone who knew nothing about it a week ago.
A lab pays a postdoc a full year to narrow a question that far. This narrowed it between midnight and breakfast, for the cost of the electricity.
Nobody deletes their own good-looking output. That is why the kill gate is a separate seat and not a preference the model can override.
How it's wired is in the article below.
Christopher Quince shares the impact of antibiotics on the human gut microbiomes using 30Tbp worth of sequences from >6k samples collected across 20 countries #GenomesMicrobiome22#Earlhaminst#largestdiagramaward
@macroliter@BrettLindenbach Ah, now you see what happens if you do 100 cycles of PCR using bacterial rDNA primers in 10% DMSO: SARS2 coronavirus becomes a virophage parasitising the monkeypox genome, to create "coronamopoxvirus"!
NEW: Single-cell transcriptomic profiling of human blood from new-born, healthy young and old, and frail adults unveils frailty-specific immune cell characteristics with genetic signatures and regulatory molecules that underlie aging https://t.co/mSomNDcavr
@gringene_bio @schwessinger Hi @gringene_bio have you tried with SFB wash on adapted library, like it did for ligation protocol? i wonder what would be the risk if washed multiple times with SFB ..
🥳Record-breaking 59 megajoules of sustained fusion energy at world-leading UKAEA’s Joint European Torus (JET) facility. Video shows the record pulse in action. Full story https://t.co/iShCGwlV9Y
#FusionIsComing#FusionEnergy#STEM#fusion@FusionInCloseUp@iterorg @beisgovuk
Studying brain metastasis? new single-cell RNAseq paper in @CellCellPress describes a detailed catalog of the functional cell types and cellular states found in human brain metastases https://t.co/f51hTjAtSO
Genetics and genomics play critical roles in health and disease, but what’s the difference between the two? Why are both important for our health? Learn more in our genetics vs. genomics fact sheet! https://t.co/U9FenuHitY