Personal website of Gwern Branwen (writer, self-experimenter, and programmer): topics: psychology, statistics, technology, deep learning, anime. This index page is a categorized list of Gwern.net pages.
Visit gwern.net → RSS feedOn Hacker News Nov 2011 – Sep 2026
Life, Learning & Writing 35Society, Policy & Culture 23Science & Space 23AI & Machine Learning 14Tools & Productivity 13Tech Industry & Business 12
Critics of AI risk suggest diminishing returns to computing (formalized asymptotically) means AI will be weak; this argument relies on a large number of questionable premises and ignoring additional resources, constant factors, and nonlinear returns to small intelligence advantages, and is highly unlikely.
In typography/design, ‘sidenotes’ place footnotes/endnotes in the margins for easier reading. I discuss design choices, HTML implementations and their pros/cons.
Observations on Frank Herbert’s 𝐷𝑢𝑛𝑒 series—the notes are probably overstated, the Butlerian Jihad was 𝑛𝑜𝑡 a robot uprising seeking to exterminate humanity, and precognition/prescience in the Dune universe appears to work backwards allowing for retro-causality and stable time-loops as exemplified by Leto II’s Golden…
How do you encrypt a file such that it can be decrypted after a date, but not before? Use serial computations for proof-of-work using successive squaring, chained hashes, or witness encryption on blockchains.
Exercise in pointing out the obvious: Greenland has been, is, and will be indefinitely, a white elephant. Denmark turned down 100m USD from the USA in 1946; I discuss how this was a bad idea—America got what it needed anyway while Denmark kept control of a loser.
Correlations are oft interpreted as evidence for causation; this is oft falsified; do causal graphs explain why this is so common, because the number of possible indirect paths greatly exceeds the direct paths necessary for useful manipulation?
Some anomalies/questions which are not necessarily important, but do puzzle me or where I find existing explanations to be unsatisfying.
What are the true limits to motivation?
Misc thoughts, memories, proto-essays, musings, etc.
Mathematical mistake/error-rates limit our understanding of rare risks and ability to defend against them.
AI folklore tells a story about a neural network trained to detect tanks which instead learned to detect time of day; investigating, this probably never happened.
Video games are art, but a strange art: their essence is transformation of the player, not description to the player. This makes meaningful criticism nearly impossible—you can point at the moon, but it’s not the moon, and once someone sees it, they no longer need the pointing.
Iron Law of Bureaucracy: the downwards deletionism spiral discourages contribution and is how Wikipedia will die.
Does DNB increase IQ? What factors affect the studies? Probably not: gains are driven by studies with weakest methodology like apathetic control groups.
On GPT-3: meta-learning, scaling, implications, and deep theory. The scaling hypothesis: neural nets absorb data & compute, generalizing and becoming more Bayesian as problems get harder, manifesting new abilities even at trivial-by-global-standards-scale. The deep learning revolution has begun as foretold.
Fictional short story about Clippy & AI hard takeoff scenarios grounded in contemporary ML scaling, self-supervised learning, reinforcement learning, and meta-learning research literature.
Archiving the Web, because nothing lasts forever: statistics, online archive services, extracting URLs automatically from browsers, and creating a daemon to regularly back up URLs to multiple sources.
Effects, health concerns, suppliers, prices & rational ordering.
A catalogue of software constructs, languages, or APIs which are unexpectedly Turing-complete; implications for security and reliability.
A classic pattern in technology economics, identified by Joel Spolsky, is layers of the stack attempting to become monopolies while turning other layers into perfectly-competitive markets which are commoditized, in order to harvest most of the consumer surplus; discussion and examples.
On desirable correlates of intelligence
A cost-benefit analysis of the marginal cost of IVF-based embryo selection for intelligence and other traits with 2016–2017 state-of-the-art.
On GPT-3: meta-learning, scaling, implications, and deep theory. The scaling hypothesis: neural nets absorb data & compute, generalizing and becoming more Bayesian as problems get harder, manifesting new abilities even at trivial-by-global-standards-scale. The deep learning revolution has begun as foretold.