All for One: LLMs Solve Mental Math at the Last Token With Information Transferred From Other TokensThis paper investigates how large language models (LLMs) solve mental math problems. It proposes that meaningful computation occurs late in the network (in terms of layer depth) and primarily at the last token, receiving information from other tokens in specific middle layers. The authors introduce techniques (CAMA and ABP) to identify an 'All-for-One' subgraph responsible for this behavior, demonstrating its sufficiency and necessity for high performance across various models and input styles.
Ancient DNA points to the roots of Uralic languages in Yakutia, far east of the Ural Mountains. The genetic trail traces a remarkable prehistoric migration that reshaped Eurasia’s linguistic landscape.
A new study using ancient DNA has revealed that the origins of the Uralic language family (including Hungarian, Finnish, and Estonian) lie further east than previously thought – specifically in Yakutia, northeastern Siberia, around 4,500 years ago. Researchers, led by Harvard scientists, analyzed genomes from Siberia and across Eurasia, tracing a genetic signal westward. This challenges the long-held belief that the Uralic homeland was near the Ural Mountains.
The study links the spread of Uralic languages to the Seima-Turbino phenomenon – a period of advanced bronze-casting and cultural exchange – and shows interaction with the Yamnaya culture (associated with the spread of Indo-European languages). While modern Uralic-speaking populations show varying degrees of Yakutia ancestry (with Hungarians having the least), the ancient DNA confirms a Siberian origin for the language family. The research also sheds light on the origins of the Yeniseian language family and potentially supports a connection between Yeniseian and North American Na-Dene languages.
A new study using ancient DNA has revealed that the origins of the Uralic language family (including Hungarian, Finnish, and Estonian) lie further east than previously thought – specifically in Yakutia, northeastern Siberia, around 4,500 years ago. Researchers, led by Harvard scientists, analyzed genomes from Siberia and across Eurasia, tracing a genetic signal westward. This challenges the long-held belief that the Uralic homeland was near the Ural Mountains.
The study links the spread of Uralic languages to the Seima-Turbino phenomenon – a period of advanced bronze-casting and cultural exchange – and shows interaction with the Yamnaya culture (associated with the spread of Indo-European languages). While modern Uralic-speaking populations show varying degrees of Yakutia ancestry (with Hungarians having the least), the ancient DNA confirms a Siberian origin for the language family. The research also sheds light on the origins of the Yeniseian language family and potentially supports a connection between Yeniseian and North American Na-Dene languages.
The article discusses a noticeable linguistic pattern emerging in AI-generated text – the negation structure 'it's not X, it's Y' – and how it's even appearing in spoken language. It explores the implications of this pattern for identifying AI-generated content and the potential impact on human writing and communication.
A new study reveals that wild chimpanzees combine vocal calls in structured ways to create new or more specific meanings, echoing key features of human language. Researchers documented over 12,000 calls from chimpanzees in Côte d’Ivoire and found that they formed at least 16 different two-call combinations.
A collection of 23 maps and charts illustrating various aspects of language, including origins, distribution, diversity, and evolution, with a focus on English and global patterns.
New genetic research suggests that humans first developed language around 135,000 years ago, with its widespread social use around 100,000 years ago. This study, using data from 15 genetic studies, indicates that language likely began as a cognitive system before becoming crucial for social communication.
A new genomic study suggests that the capacity for human language emerged at least 135,000 years ago, coinciding with the initial geographic divergence of Homo sapiens. This capacity likely transitioned into social use around 100,000 years ago, as indicated by symbolic activities in the archaeological record.
A pair of landmark studies published in Nature have identified the origin of the Indo-European family of over 400 languages, spoken by more than 40% of the world's population, to the Caucasus Lower Volga people in present-day Russia around 6,500 years ago.
Harvard researchers traced the origins of the vast Indo-European language family to the Caucasus-Lower Volga region, identifying the ancestral population known as the Yamnaya, who appeared around 3300 BCE and spread from Hungary to western China.
