corpus
A large collection of text or code used as training data for an AI model.
As defined in Paul Graham
Where it comes up · 5
- 181 | Peter Dodds on Quantifying the Shape of Stories3 claims
- Gaurav Kapadia2 claims
- Edward Gibson2 claims
- Paul Graham1 claim
- Jimmy Wales1 claim
Most important claims · 9 of 9Search all claims for “corpus” →
- Machine translation historically was only useful for economically important languages because of the corpora used, but new large-language-model methods have dramatically improved translation for smaller languages like Zulu, which will let small-language communities produce shareable articles and accelerate the largely invisible growth of Wikipedia in developing-world languages.Jimmy Wales
- Across roughly 60 languages with parsed dependency corpora, the actual dependency lengths in real sentences are consistently shorter than in random reorderings of the same words, providing quantitative cross-linguistic evidence for dependency length minimization.Edward Gibson
- Gibson's biggest difference from Chomsky is methodological: Gibson gathers quantitative data through experiments and corpus analysis, while Chomsky relies on thought experiments and his own intuitions and has never been associated with any experiment or corpus work.Edward Gibson
- Across 10 major languages and 24 corpora, human language exhibits a Pollyanna principle: there are more positive words than negative words, which is surprising given how easily negativity comes to mind, but reflects the fact that society holds together through countless small positive interactions—language being our great social technology.181 | Peter Dodds on Quantifying the Shape of Stories
- AI and language systems have raced ahead into the 'create' category of science before we adequately completed the 'describe' and 'explain' stages, leading to systems used for decisions about juries or parole that turned out to be deeply racist; Dodds is relieved that people are now going back to examine the corpora these systems were built from.181 | Peter Dodds on Quantifying the Shape of Stories
- You cannot bootstrap a better programming language by having AI write code, training other AIs on it, and converging — a recent research paper found that training an AI on the output of AIs converges on crap; you must have a large initial corpus of human-written examples in the language.Paul Graham
- AI will deliver leverage to investing within the next year by handling much of the analysts' time-consuming initial company evaluations, freeing analysts and partners to spend most of their energy applying judgment and taste to the corpus of facts, while also dramatically changing tooling away from the Bloomberg/Excel desktop and easing the heavy legal, compliance, operations, and tax burdens of running an investment firm.Gaurav Kapadia
- XN runs concentrated public-markets investing—rarely more than 10-15 positions—because there are very few genuinely good ideas, so capital is concentrated on the best ones; combined with a long holding period, this means adding only one to three ideas a year and keeping an extremely high bar.Gaurav Kapadia
- The collective English text in Google Books is full of science—medical and scientific writing—and the 20th century is dominated by the rise of science, visible in details like the word 'Figure' (capital F) and 'et al' rising sharply, reflecting the exponential growth of scientific publishing.181 | Peter Dodds on Quantifying the Shape of Stories