A useful mental model for an LLM is a giant matrix where each row is a possible prompt and columns represent next-token probabilities. This matrix is impossibly large but also extremely sparse, as most token combinations are gibberish. The LLM's job is to efficiently compress and approximate this matrix.

The brain's hardware limitations, like slow and stochastic neurons, may actually be advantages. These properties seem perfectly suited for probabilistic inference algorithms that rely on sampling—a task that requires explicit, computationally-intensive random number generation in digital systems. Hardware and algorithm are likely co-designed.

Contrary to fears, interpretability techniques for Transformers seem to work well on new architectures like Mamba and Mixture-of-Experts. These architectures may even offer novel "affordances," such as interpretable routing paths in MoEs, that could make understanding models easier, not harder.

While both humans and LLMs perform Bayesian updating, humans possess a critical additional capability: causal simulation. When a pen is thrown, a human simulates its trajectory to dodge it—a causal intervention. LLMs are stuck at the level of correlation and cannot perform these essential simulations.

Contrary to the belief that memorization requires multiple training epochs, large language models demonstrate the capacity to perfectly recall specific information after seeing it only once. This surprising phenomenon highlights how understudied the information theory behind LLMs still is.

Google's Titans architecture for LLMs mimics human memory by applying Claude Shannon's information theory. It scans vast data streams and identifies "surprise"—statistically unexpected or rare information relative to its training data. This novel data is then prioritized for long-term memory, preventing clutter from irrelevant information.

To ensure scientific validity and mitigate the risk of AI hallucinations, a hybrid approach is most effective. By combining AI's pattern-matching capabilities with traditional physics-based simulation methods, researchers can create a feedback loop where one system validates the other, increasing confidence in the final results.

The 2017 introduction of "transformers" revolutionized AI. Instead of being trained on the specific meaning of each word, models began learning the contextual relationships between words. This allowed AI to predict the next word in a sequence without needing a formal dictionary, leading to more generalist capabilities.

Research shows it's possible to distinguish and remove model weights used for memorizing facts versus those for general reasoning. Surprisingly, pruning these memorization weights can improve a model's performance on some reasoning tasks, suggesting a path toward creating more efficient, focused AI reasoners.