Research snapshot: 8 October 2026; corrections checked 9 October. Later evidence is dated explicitly rather than silently added to the original cutoff. Publication dates below distinguish first o…
A Mechanistic Fallibilist
Dossier
The sources and notes behind the essay.
/
Use quotes for an exact phrase.
All research notes
About these notes
Overview
Fallibilism treats our knowledge claims as open to correction. “Mechanistic fallibilist” is my framing for combining explanations with tests that can expose their failure. Stanford Encyclopedia o…
TasteVal reports a 2.3× compute multiplier for Opus 5.5, which was evaluated on seven of eight tasks after refusing one, and different fitted trends for compute efficiency and final performance.…
A-Lab's corrected synthesis count, Vals's computational candidates, and Periodic's lab-data training results concern different stages of discovery. Together they motivate my argument about the ex…
The essay argues that people should remain accountable for scientific objectives. That commitment does not depend on proving that machines will never choose useful goals. Schulman offers an expec…
Automated laboratories
Dyna’s September 29, 2026 report defines a workflow around infrequent human handoffs, with a person returning hours later. The scientist leaving the lab is an interpretation of this company frami…
Anthropic’s life-sciences head Eric Kauderer-Abrams confirmed its wet lab and emphasized physical experiments. Reuters interview, reported September 18, 2026.
Ginkgo announced on September 15, 2026 that it would build an autonomous lab at Novo’s Waltham R&D site. CEO Jason Kelly described around-the-clock experiments freeing scientists for analysis and…
The story draws on reported classes of failure, not a verified incident corresponding to every scene. Reported laboratory operations, assay failure classes.
June 16, 2026 online; July/August print issue
Scientific American’s web report, The lab never sleeps. Can the science keep up?, appeared in print as The Lab of the Future Runs Itself, volume 335, issue 1. It describes overnight operation, al…
September 24, 2026 preprint; September 25 company account
The catalyst campaign reports 2,942 catalysts across 53 material systems and 26 elements on a more-than-90%-automated, human-supervised platform. Its iridium- and ruthenium-free palladium-oxide c…
April 23, 2026
Medra announced Lab 001 and describes natural-language instruction of its platform. The announcement supports the company’s stated system and plans; a claim to be the largest facility would requi…
The Nobel record establishes the transposition discovery and sole 1983 award. Nathaniel Comfort’s 1999 archival study distinguishes acceptance of transposition from controversy over the broader c…
Comfort’s archival study locates the controversy in her broader theory of genetic control, while documenting early acceptance of transposition. Howard Green, who knew her personally, recalls low…
Her maize experiments revealed specific interactions between genetic elements that altered gene activity and inherited pigmentation patterns. Movement alone did not describe those control relatio…
August 17, 2022 and September 1, 2026
Conventional single-cell RNA-seq is destructive, but Live-seq sampled living cells in 2022. Live-cell transcriptomics with engineered virus-like particles was published online in Cell on Septembe…
April 25
Non-terminal vesicular export offers another route to nondestructive transcriptomic sampling. It reinforces the need to qualify the destructive-measurement premise, without making all biological…
Experimental timescales
TasteVal v2 distinguishes more sequential feedback rounds from more experiments per round; its runs use one H100 and one experiment at a time. TasteVal §2.5.1. Interpretation: a lab step that mus…
18 of 300