• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
Technology
Report

Claude and BootLoops effort reportedly produced 36 manuscripts across 18 fields in three months

Anthropic says physicist Matthew Schwartz built a calculation toolkit that helped Claude find connections across scientific fields.

CH
DS
2 Sources, 1d ago, first seen 1d ago

TLDR

Anthropic says Schwartz used a toolkit for exact calculations, while domain experts helped steer Claude toward interesting questions. Posts describing his Claude and BootLoops effort say it explored roughly 400 candidate problems and produced 36 manuscripts across 18 fields with 19 coauthors in three months.

Combined views

151.5K

2 Sources, first seen 1d ago

1.9K likes44 comments694 saves234 reposts
Featured Source

Combined views

151.5K

2 Sources, first seen 1d ago

1.9K likes44 comments694 saves234 reposts

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

2 Sources

@choblin29HOLY MOLY: Anthropic just published what may be one of the wildest examples yet of AI-accelerated science. Physicist Matthew Schwartz says his Claude + BootLoops setup produced 36 MANUSCRIPTS across 18 fields with 19 coauthors in just THREE MONTHS, after exploring ~400 candidate problems. >solved an ecology equation nobody had been able to scale for 20 years >analyzed 5.7 BILLION pairs of mutations and found evidence for gene conversion >ported 4,452 economics papers, ~30,000 routines, and checked essentially every validatable number >built a word-stress database covering 6,072 languages from 160,000 phonology works And the weirdest part: Schwartz says Claude was often technically correct but scientifically uninteresting until domain experts stepped in and told it what questions actually mattered.1d
@Dr_Singularityhuge news from Anthropic / amazing acceleration AI is starting to do years of science in days 👀 Anthropic published an extraordinary account of Prof. Matthew Schwartz using Claude + his BootLoops system across 18 scientific fields. In just 3 months, the effort produced 36 manuscripts with 19 coauthors, selected from roughly 400 candidate problems. Some of the standout results: - Claude reproduced results that had taken Schwartz weeks to code in about 20 minutes, then helped generalize the method to harder classes of physics problems. - The system completed 30 difficult Feynman integrals, including 15 that had apparently never been computed before. - In ecology, it solved an equation that had resisted scalable computation for about 20 years, then helped build a better predictive model with domain experts. - In population genetics, it analyzed 5.7 billion pairs of nearby mutations and found evidence for gene conversion that could matter for how genetic datasets are interpreted. - In economics, Claude helped port and check replication packages from 4,452 papers, covering around 30,000 routines. - In linguistics, it helped create a word stress database spanning 6,072 languages and a bibliography of 160,000 phonology works. - It also contributed to work in Earth science, genomics, cosmology, statistics, phylogenetics, sunspots, and mathematical physics. Now imagine what the next generation of models will do. Science is about to move much, much faster.1d
    • HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI

    2 Sources

    @choblin29HOLY MOLY: Anthropic just published what may be one of the wildest examples yet of AI-accelerated science. Physicist Matthew Schwartz says his Claude + BootLoops setup produced 36 MANUSCRIPTS across 18 fields with 19 coauthors in just THREE MONTHS, after exploring ~400 candidate problems. >solved an ecology equation nobody had been able to scale for 20 years >analyzed 5.7 BILLION pairs of mutations and found evidence for gene conversion >ported 4,452 economics papers, ~30,000 routines, and checked essentially every validatable number >built a word-stress database covering 6,072 languages from 160,000 phonology works And the weirdest part: Schwartz says Claude was often technically correct but scientifically uninteresting until domain experts stepped in and told it what questions actually mattered.1d
    @Dr_Singularityhuge news from Anthropic / amazing acceleration AI is starting to do years of science in days 👀 Anthropic published an extraordinary account of Prof. Matthew Schwartz using Claude + his BootLoops system across 18 scientific fields. In just 3 months, the effort produced 36 manuscripts with 19 coauthors, selected from roughly 400 candidate problems. Some of the standout results: - Claude reproduced results that had taken Schwartz weeks to code in about 20 minutes, then helped generalize the method to harder classes of physics problems. - The system completed 30 difficult Feynman integrals, including 15 that had apparently never been computed before. - In ecology, it solved an equation that had resisted scalable computation for about 20 years, then helped build a better predictive model with domain experts. - In population genetics, it analyzed 5.7 billion pairs of nearby mutations and found evidence for gene conversion that could matter for how genetic datasets are interpreted. - In economics, Claude helped port and check replication packages from 4,452 papers, covering around 30,000 routines. - In linguistics, it helped create a word stress database spanning 6,072 languages and a bibliography of 160,000 phonology works. - It also contributed to work in Earth science, genomics, cosmology, statistics, phylogenetics, sunspots, and mathematical physics. Now imagine what the next generation of models will do. Science is about to move much, much faster.1d
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet