AI Summary

Researchers at Sapient claim to have successfully trained a foundation model from scratch for a relatively low cost of around $1,500. This achievement is significant because it challenges the conventional wisdom that training a foundation model requires significant resources and internet-scale data, which can be prohibitively expensive for many enterprises. The researchers used a Hierarchical Recurrent Model (HRM) to replace standard Transformers, resulting in a more sample-efficient approach. The implications of this development are that it may be possible for smaller organizations or those with limited budgets to develop and train their own foundation models, potentially leading to more customized and effective AI solutions.

AI summaries can be wrong sometimes—always verify important details using the source article.