The article introduces **talkie-1930-13b-base**, a **13B “vintage” language model** trained on **260 billion tokens of pre-1931 English text**. Vintage LMs are designed to simulate conversations with
The primary challenge facing AI today is poor generalization rather than issues of scale or architecture. Current AI models struggle to generalize as effectively as humans, making sample efficiency an
Examining the Case for AGI
news
amardashehu.substack.com
Ehzd1ZUbifpck9XyJf9d/9rX0i3KTg3rh/c4Kceg1iI=
Feb 18
In her essay, Amarda Shehu critiques a recent article in *Nature* that claims AI has achieved human-level intelligence, arguing that the assertions within lack rigorous empirical support. Shehu emphas