"In a mindless age, every insight takes on the character of a lethal weapon." - M. McLuhan to E. Pound, letter, 1951.

Só pra lembrar que humilhação não é o Brasil ter perdido da Noruega. Humilhação é perder pra um país de 5,5 milhões de habitantes que descobriu petróleo, olhou pra riqueza e pensou nos netos. Criaram um fundo soberano que hoje vale US$ 2,2 trilhões, quase R$ 11 trilhões, dinheiro que daria R$ 1,8 milhão pra cada norueguês. A Noruega é dona de 1,5% de todas as ações do planeta. O Brasil descobriu o pré-sal e transformou em emenda secreta e refinaria superfaturada. Nossos jogadores não devem nada aos deles. Nossa classe política deve tudo. O que a gente tem que importar da Noruega não é futebol, é vergonha na cara.

Extremely grateful to @MalloryMcMorrow, who inspired a national race to the top on AI safety policy. She practically dared @LeadingFutureAI to oppose her and they didn't, which shows just how toxic industry money has become in Democratic primaries.

I like this: a global village, with all of the vices of the globe and the village and none of their virtues.

The onslaught of papers doesn't take a holiday apparently. This will be a must-read.

Karp's jeremiad and the @quxiaoyin's takes don't make sense as generalized critiques. These models are general-purpose intelligence in software form. There will *always* be a market for 99th percentile intelligence because the returns to marginal intelligence are immense. Why do finance, law, and tech firms pay so much for Harvard, MIT, Stanford, and Berkeley grads? Because their exceptional intelligence is a key input into the production of premium products. Many diverse development and business models will rely on frontier AI systems because generalized intelligence will be key to their product development/products/moats. Their primary alpha will be in faster production of better products and products that effectively operate generalized intelligence. For these firms, high API fees are worthwhile, but they enable delivery of substantial value-add to customers by making development of core products cheaper or providing products that provide substantial performance improvement over competitors/what frontier models supply on their own. Others will be able to rely on non-frontier open-source models (or won't heavily leverage AI systems at all), because their alpha is primarily derived from human artisanship, brand value, etc. The economy is huge: there's clearly sundry use-cases in each. Chatbot integrations for Amazon probably don't require generalizable frontier models; future e-commerce paradigms may, however. Either way, the idea that some companies' idiosyncratic preference for local hosting of customized open-weight models dooms the market for API-accessible frontier intelligence is misguided. It's also not clear why such companies would substantially benefit from accessing these models via Palantir's pricey software, which in any case is mostly hosted in AWS -- an external dependency per Karp's logic?