arXiv:2601.07968v1 Announce Type: new Abstract: We study the theoretical problem of synthesizing multiple DNA strands under spatial constraints, motivated by large-scale DNA synthesis technologies. In this setting, strands are arranged in an array and synthesized according to a fixed global synthesis sequence, with the restriction that at most one strand per row may be synthesized in any synthesis cycle. We focus on the basic case of two strands in a single row and analyze the expected completion time […]

Ver mais

This paper argues that transformers are being overused as universal execution engines. I propose a meaning-first execution framework that decouples semantic proposal from model execution, allowing inference to be conditionally invoked only when needed. The result is that a large fraction of transformer calls can be skipped without changing correctness on invoked cases, suggesting many current efficiency limits are architectural rather than model-intrinsic. The work is model-agnostic and sits above existing transformers. Feedback welcome, especially around routing guarantees […]

Ver mais

arXiv:2601.08998v1 Announce Type: new Abstract: Flaky tests are a common problem in software testing. They produce inconsistent results when executed multiple times on the same code, invalidating the assumption that a test failure indicates a software defect. Recent work on LLM-based test generation has identified flakiness as a potential problem with generated tests. However, its prevalence and underlying causes are unclear. We examined the flakiness of LLM-generated tests in the context of four relational database management systems: SAP […]

Ver mais

I recently started doing software development work to expand and integrate AI systems. Here are some of the biggest surprises I found along the way. AI Developers Love Calling Everything “Fine Tuning” When studying for the AI-102 exam, I thought fine tuning strictly referred to additional training of a model to include additional data (which can be fed into the system in different ways). When actually working on an AI project, what I learned was developers (and architects) sprinkle […]

Ver mais

GitHub: Engram: https://github.com/deepseek-ai/Engram arXiv:2601.07372 [cs.CL]: https://arxiv.org/abs/2601.07372 “While Mixture-of-Experts (MoE) scales capacity via conditional computation, Transformers lack a native primitive for knowledge lookup, forcing them to inefficiently simulate retrieval through computation. To address this, we introduce conditional memory as a complementary sparsity axis, instantiated via Engram, a module that modernizes classic N-gram embedding for O(1) lookup. By formulating the Sparsity Allocation problem, we uncover a U-shaped scaling law that optimizes the trade-off between neural computation (MoE) and static memory […]

Ver mais

arXiv:2601.09018v1 Announce Type: new Abstract: Across engineering and scientific domains, traditional deep learning (TDL) models perform well when training and test data share the same distribution. However, the dynamic nature of real-world data, broadly termed textit{data shift}, renders TDL models prone to rapid performance degradation, requiring costly relabeling and inefficient retraining. Meta-learning, which enables models to adapt quickly to new data with few examples, offers a promising alternative for mitigating these challenges. Here, we systematically compare TDL with […]

Ver mais

Jürgen Schmidhuber shares his insights on the evolution and future of AI. He discusses the development of self-improving AI systems, the concept of artificial curiosity, and the potential for machines to achieve creativity akin to humans. Schmidhuber also explores the implications of advanced AI on society, emphasizing the importance of aligning AI goals with human values to ensure beneficial outcomes. He envisions a future where AI contributes positively to various fields, including science, art, and technology.

Ver mais

arXiv:2404.17592v2 Announce Type: replace-cross Abstract: As e-commerce expands, delivering real-time personalized recommendations from vast catalogs poses a critical challenge for retail platforms. Maximizing revenue requires careful consideration of both individual customer characteristics and available item features to continuously optimize assortments over time. In this paper, we consider the dynamic assortment problem with dual contexts — user and item features. In high-dimensional scenarios, the quadratic growth of dimensions complicates computation and estimation. To tackle this challenge, we introduce a […]

Ver mais

Humans don’t start their thinking from scratch every second. As you read this essay, you understand each word based on your understanding of previous words. You don’t throw everything away and start thinking from scratch again. Your thoughts have persistence. Traditional neural networks can’t do this, and it seems like a major shortcoming. For example, imagine you want to classify what kind of event is happening at every point in a movie. It’s unclear how a traditional neural […]

Ver mais