
Physicist, Immigrant, Pilot, Dad. Former Caltech, Hyperloop, NASA JPL. Once considered pretty smart for an organic. Founder @terraformindies. Hiring!
NASAdena
Joined June 2012
- Tweets42.9K
- Following5.2K
- Followers72.5K
- Likes128K
Pinned Tweet
In my ongoing quest to find the limits of Opus 5.5, I directed it to audit my old scientific field, theoretical astrophysics. We indexed about 150,000 papers, audited 50,000 for issues, and did progressively more thorough checks on thousands more. We found nearly 1000 major errors that got cited many times, most are still uncorrected. This survey enabled me to estimate the efficacy of peer review for correcting errors (not very good) and to identify dozens of open questions that anyone with spare Claude Max tokens could answer today (.md context files included).
Bottom line: AI can speed up quantitative science by >1000x with no further model improvement required. But I'm seeing no evidence of this being pushed within universities, only academics complaining about labs dropping math proofs.
Among my fellow reformed academic high agency founders, however, several are speed running entire disciplines.
This survey has implications for how science will be run in the very near future. Hint: Academic publishing is dead. It was already a very stupid way to record and promulgate knowledge, and had been so for decades, even before all that publicly funded knowledge was locked behind paywall cartels. Now AIs can generate thousands of novel, correct papers per day, it is no longer meaningful for humans to consider a research paper a "unit of science" or to even bother much with the 99.9% of science which, up to this point, was necessary drudgery. Time to discover something like General Relativity or Evolution every week. LFG!!
Enjoy: science.caseyhandmer.com/phy…
Casey Handmer retweeted
If you’re reading this, you’re in the top ~0.005% on earth for understanding how many legacy systems are about to get routed around.
In my ongoing quest to find the limits of Opus 5.5, I directed it to audit my old scientific field, theoretical astrophysics. We indexed about 150,000 papers, audited 50,000 for issues, and did progressively more thorough checks on thousands more. We found nearly 1000 major errors that got cited many times, most are still uncorrected. This survey enabled me to estimate the efficacy of peer review for correcting errors (not very good) and to identify dozens of open questions that anyone with spare Claude Max tokens could answer today (.md context files included).
Bottom line: AI can speed up quantitative science by >1000x with no further model improvement required. But I'm seeing no evidence of this being pushed within universities, only academics complaining about labs dropping math proofs.
Among my fellow reformed academic high agency founders, however, several are speed running entire disciplines.
This survey has implications for how science will be run in the very near future. Hint: Academic publishing is dead. It was already a very stupid way to record and promulgate knowledge, and had been so for decades, even before all that publicly funded knowledge was locked behind paywall cartels. Now AIs can generate thousands of novel, correct papers per day, it is no longer meaningful for humans to consider a research paper a "unit of science" or to even bother much with the 99.9% of science which, up to this point, was necessary drudgery. Time to discover something like General Relativity or Evolution every week. LFG!!
Enjoy: science.caseyhandmer.com/phy…
It's been terrific to watch Caleb and Molten bring this capability from sketch to commercial reality. The future is unimaginably bright!
Over the last 18 months, the @Molten_Ind team has literally worked day and night to build and commission the world's first plant to turn natural gas into battery-grade graphite and clean hydrogen.
Molten graphite delivers state-of-the-art performance, verified by leading global battery producers, with 3x less electricity and zero process emissions.
100% US-made and operating now at a modular commercial scale.
@TechCrunch and the brilliant @tdechant have the story below
Hearing the @united CEO Scott Kirby on @CheekyPint telling @collision that the primary constraint on commercial aviation in the US is ATC slots drives me crazy. There are at most about 5000 flights occurring at any one time over the entire US. There are about 30 usable flight levels in alpha airspace, and you can deconflict 100x more flights than this with a single toothbrush CPU (FAA.caseyhandmer.com).
In terms of physical space, there's enough room to fly every man woman and child in the US simultaneously. Look up. Does the sky seem crowded?
There were plans to automate ATC using the first Iridium constellation more than 30 years ago.
At some point we have to have a serious conversation about rent takings by obsolete incumbents that are blocking entire percentage points of economic growth.
Incidentally this is how Chandra worked in physics, back when our best AIs were obsessive humans who worked 5000 hours per year.
Just confirmed that today we completed integration testing and produced e-methanol from sunlight and air with our own solar on our own land.
Five years to instantiate the Jeffersonian dream of rugged individualism and self sufficiency. "Those who labour in the earth are the chosen people of God."
An enduring monument to the brilliance and dedication of our incredible team.
Next up: The trickle becomes an unstoppable flood. We are determined to banish energy scarcity forever.
If someone at one of the leading labs wants to sponsor O($100,000) of tokens, I can complete this particular study (audit essentially all the papers to the limits of current models) OR repeat this analysis on essentially all fields of science (statistically significant detailed error sampling of key papers in a field). Or you can. It's not that hard.
In my ongoing quest to find the limits of Opus 5.5, I directed it to audit my old scientific field, theoretical astrophysics. We indexed about 150,000 papers, audited 50,000 for issues, and did progressively more thorough checks on thousands more. We found nearly 1000 major errors that got cited many times, most are still uncorrected. This survey enabled me to estimate the efficacy of peer review for correcting errors (not very good) and to identify dozens of open questions that anyone with spare Claude Max tokens could answer today (.md context files included).
Bottom line: AI can speed up quantitative science by >1000x with no further model improvement required. But I'm seeing no evidence of this being pushed within universities, only academics complaining about labs dropping math proofs.
Among my fellow reformed academic high agency founders, however, several are speed running entire disciplines.
This survey has implications for how science will be run in the very near future. Hint: Academic publishing is dead. It was already a very stupid way to record and promulgate knowledge, and had been so for decades, even before all that publicly funded knowledge was locked behind paywall cartels. Now AIs can generate thousands of novel, correct papers per day, it is no longer meaningful for humans to consider a research paper a "unit of science" or to even bother much with the 99.9% of science which, up to this point, was necessary drudgery. Time to discover something like General Relativity or Evolution every week. LFG!!
Enjoy: science.caseyhandmer.com/phy…
Casey Handmer retweeted
It's crazy that Europe has issues with depending on foreign adversaries for fuel given that it has such vast reserves of oil that it doesn't exploit due to politics.
Can't wait for the NIH to peel off 0.01% of their budget to audit the efficacy of the last decade of grants that spent ~$100b on neurodegenerative disease research with zero effective drugs being found...
In my ongoing quest to find the limits of Opus 5.5, I directed it to audit my old scientific field, theoretical astrophysics. We indexed about 150,000 papers, audited 50,000 for issues, and did progressively more thorough checks on thousands more. We found nearly 1000 major errors that got cited many times, most are still uncorrected. This survey enabled me to estimate the efficacy of peer review for correcting errors (not very good) and to identify dozens of open questions that anyone with spare Claude Max tokens could answer today (.md context files included).
Bottom line: AI can speed up quantitative science by >1000x with no further model improvement required. But I'm seeing no evidence of this being pushed within universities, only academics complaining about labs dropping math proofs.
Among my fellow reformed academic high agency founders, however, several are speed running entire disciplines.
This survey has implications for how science will be run in the very near future. Hint: Academic publishing is dead. It was already a very stupid way to record and promulgate knowledge, and had been so for decades, even before all that publicly funded knowledge was locked behind paywall cartels. Now AIs can generate thousands of novel, correct papers per day, it is no longer meaningful for humans to consider a research paper a "unit of science" or to even bother much with the 99.9% of science which, up to this point, was necessary drudgery. Time to discover something like General Relativity or Evolution every week. LFG!!
Enjoy: science.caseyhandmer.com/phy…
One AI-pilled first year grad student could blitz an entire department or field. As a result, we see the strongest sociological norms against AI use and defection in academia. Fortunately, enough grad students are neurodivergent enough that they do not care and will do it anyway.
Much of the academic world and particularly in Ireland are refusing to accept that AI is going to help every discipline go faster and with better quality. The sooner the better everybody jumps on this train and moves society forward faster, the better.
Why hasn't every university in Ireland give unlimited tokens to their students?
Casey Handmer retweeted
Replying to @stephen_winters
You can literally cap an entire field in a few days and build new knowledge structures on top that are arbitrarily mutable, legible, computable, verifiable, query-able, and inference-able.




















