#1 - Episode 1: Transformer 2.0

#1 - Episode 1: Transformer 2.0

In this episode of Brain Snacks, host Noah Harrington dives deep into what might be "Transformer 2.0" - two revolutionary papers reshaping the foundation of AI language models: • Google DeepMind's "Titans: Learning to Memorize at Test Time [https://arxiv.org/abs/2501.00663]": Revolutionizing how transformers handle memory by introducing: - Neural long-term memory modules that learn during inference - Ability to process over 2 million tokens at test time - Novel "surprise metric" for selective memory storage • Sakana AI's "Transformer²: Self-Adaptive LLMs [https://sakana.ai/transformer-squared/]": Reinventing transformer architecture through: - Self-adaptive AI that rewires its own structure - Dynamic component modification using Singular Value Decomposition - More efficient computation while maintaining performance Perfect for anyone curious about the next evolution of transformer models and how they're becoming more like human brains. No dense technical jargon - just clear, fascinating insights into the future of AI language models.