# Summary of the article:
– Transformer-based models have made great strides in NLP tasks but face difficulties in long context reasoning, multi-step inference, and numerical reasoning.
– These challenges stem from their quadratic self-attention complexity and lack of explicit memory.
## Author’s take:
Convergence Labs has unveiled the Large Memory Model (LM2) to tackle the hurdles faced by Transformer-based models in handling long context reasoning. This memory-augmented transformer architecture seems promising in addressing the deficiencies highlighted, ushering in a new era of advancements in NLP capabilities.
Click here for the original article.