The recent drop in # shows, that many people who reacted to the #DeepSeeks alleged ability to need less compute power don't understand, that #AI is and always has been about computation efficiency.

My secret passion is getting to learn of “Apple expert users” that the feature I need is not there because I don’t actually want that on the Apple support forum.

Concepts for working with little specific data to a given task have first gained significant relevance for AI with the first #GPT models. Their relevance has ever grown since. Thanks to architectures like #TabPFN it is also possible, to apply #FewShotLearning to applications, where precision is key.

The new training schedules used a #Deepseek #R1 show how #llms can learn to solve tasks by themself! Especially coding models will rapidly get better in the next months and years!