
DeepSeek Engram: Conditional Memory via Scalable Lookup
14/1/2026 | 39 mins.
This episode introduces Engram, a new architectural module that integrates conditional memory into Large Language Models to handle static knowledge more efficiently. Traditional models often waste computational depth simulating memory retrieval, but Engram uses $N$-gram lookup tables to retrieve information in constant time. By balancing this memory module with Mixture-of-Experts (MoE) computation, the authors discovered a U-shaped scaling law that optimizes performance for a fixed parameter budget. Experimental results show that Engram-enhanced models significantly outperform standard MoE baselines in general reasoning, coding, and long-context tasks. Mechanistically, the module functions by offloading local pattern reconstruction from early layers, effectively increasing the model's functional depth. Furthermore, its deterministic retrieval allows for efficient host memory offloading, enabling massive parameter scaling with minimal impact on inference speed

End-to-End Test-Time Training for Long Context
02/1/2026 | 34 mins.
This episode introduces TTT-E2E, a novel method for long-context language modeling that treats context processing as a continual learning problem rather than a structural design challenge. Instead of relying on traditional attention mechanisms that slow down as text grows, the model compresses information into its internal weights by learning at test time through next-token prediction. By utilizing meta-learning during the initial training phase, the authors optimize the model's ability to update itself efficiently on new sequences. Experiments on 3B-parameter models demonstrate that this approach maintains the performance of full-attention Transformers while achieving 2.7× faster inference at 128K context lengths. Ultimately, the method offers a hardware-efficient alternative to RNNs and Transformers by providing constant inference latency without sacrificing the ability to leverage massive amounts of data

Computational intelligence in data-driven
01/1/2026 | 40 mins.
This episode about Cris Doloc’s book explores the intersection of computational intelligence and quantitative finance, emphasizing how data-driven paradigms are revolutionizing modern trading. The author distinguishes between the theoretical hype of artificial intelligence and the practical utility of algorithmic learning, advocating for a rigorous engineering approach to market analysis. By examining high-frequency data and market microstructure, the text illustrates how machines can optimize trade execution and predict price dynamics more effectively than traditional models. Detailed case studies on portfolio management, market making, and derivatives valuation provide a blueprint for applying machine learning to complex financial problems. Ultimately, the work highlights a paradigm shift toward "algorithmic culture," where data inference and hardware acceleration replace rigid mathematical assumptions. Use of these advanced technologies aims to enhance risk management and decision-making across the digital economy

Notes on Complexity
01/1/2026 | 53 mins.
In this episode, a pathologist explores complexity theory to bridge the gap between scientific materialism and spiritual existence. By examining systems ranging from ant colonies to human cells, the author illustrates how simple, local interactions generate unpredictable emergent behaviors. The narrative highlights complementarity, arguing that the universe is a holarchy where the same entity appears as a solid body, a dance of cells, or a cloud of atoms depending on the observer’s scale. Limitations in empirical science and formal logic, exemplified by quantum mechanics and Gödel’s incompleteness theorems, suggest that reality cannot be fully captured by math alone. Ultimately, the author proposes fundamental awareness, a model where consciousness is the primary fabric of the universe rather than a mere byproduct of the brain. This perspective integrates modern physics with ancient mystical traditions to suggest we are all interconnected expressions of a single, living whole

Complexity and the Econominy
01/1/2026 | 31 mins.
This episode introduces complexity economics, a framework that views the economy as an evolving, nonequilibrium system rather than a static machine. Unlike traditional models that assume perfect rationality and steady states, this approach emphasizes how individual agents constantly adapt their strategies based on the patterns they collectively create. The research highlights positive feedbacks and increasing returns, which can lead to unpredictable outcomes like market lock-ins or sudden financial crashes. Through experiments like the El Farol bar problem and artificial stock markets, the author demonstrates how inductive reasoning and learning drive economic life. Additionally, the sources explore the evolution of technology, illustrating how new innovations emerge by combining simpler existing elements to satisfy human needs. Ultimately, the work advocates for failure-mode analysis to prevent the exploitation of policy systems, treating the economy as a living, organic process



The Gist Talk