xLSTM more expressive than transformer, Mamba: arxiv.org/abs/2603.03612 *nonlinear RNNs: sLSTM, LSTM *DPLR linear RNNs: mLSTM, RWKV, DeltaNet *Non PNC1: Mamba, Transformer “fundamental expressivity gaps between linear and nonlinear RNNs” World models require nonlinear RNNs.

Mythos is a looped transformer!? 😳 Should be a Mixture-of-Recursions (MoR) — 2× faster, controlled effort. Dense → sparse MoE was the efficiency unlock of 2023. Uniform loops → MoR is the same move for recursive transformers. Paper reading list below. 🧵

NEW paper from Apple. Interesting idea: "Attention to Mamba". The paper introduces a two-stage recipe for cross-architecture distillation from Transformers into Mamba. Naive distillation collapses teacher performance. Their trick: first distill the transformer into a

Yann LeCun says the AI industry is completely LLM pilled.

> be Yann LeCun > spend years building JEPA at Meta > company focuses on LLaMA instead > his idea stays complicated and unused > robotics plans get dropped > decides to leave and start AMI Labs > builds a much simpler version from scratch > trains it on normal hardware in just a

Si has llegado hasta aquí, claramente formas parte de los que van a ir por delante. ⭐ Comparto este tipo de cosas regularmente aquí → Alejo ¿Seguimos?

joooooooder

NORMAL PEOPLE premiered 6 years ago today.

Perdre el mòbil és fàcil. Recuperar-lo també ho pot ser si abans l’has deixat ben configurat amb aquestes quatre accions preventives 3cat.cat/3catinfo/que-c…

Me sumo a la advertencia: Sci-Hub ha pirateado más de 85 millones de artículos de investigación y ahora encima han añadido un bot que responde preguntas utilizando artículos completos y recientes. Esto es un escándalo. Dejo el enlace abajo para que sepas cómo evitarlo.

The new Banksy sculpture in London is brilliant

x.com/i/article/2047…

aftersun (2022) we still talk about you

SPLICE: Latent Diffusion over JEPA Embeddings for Conformal Time-Series Inpainting ift.tt/mVyjuPI

Adaptive Norm-Based Regularization for Neural Networks ift.tt/3KHr8FC

🚨 Karpathy acaba de soltar 40 minutos de oro sobre agentes IA en 2026. Qué aprender, qué construir y qué tirar antes de que te hunda. El 90% de las herramientas actuales no van a sobrevivir 90 días. El filtro ya está hecho. Gratis.

Language Models Interview Handbook drive.google.com/file/d/1SikCWx…