arXiv:2604.14246v1 Announce Type: new Abstract: Sparse Mixture-of-Experts (MoE) models have achieved remarkable scalability, yet they remain vulnerable to hallucinations, particularly when processing long-tail knowledge. We identify that this fragility stems from static Top-$k$ routing: routers tend to favor high-frequency patterns over rare factual associations. Consequently, “specialist experts” possessing critical long-tail knowledge are often assigned low gating scores and remain “dormant” — under-prioritized for specific tokens despite their proven causal importance on other inputs. To address this, we propose […]

Ver mais

Data-driven decision making frequently relies on predicting counterfactual outcomes. In practice, researchers commonly train counterfactual prediction models on a source dataset to inform decisions on a possibly separate target population. Conformal prediction has arisen as a popular method for producing assumption-lean prediction intervals for counterfactual outcomes that would arise under different treatment decisions in the target population of interest. However, existing methods require that every confounding factor of the treatment-outcome relationship used for training on the source data […]

Ver mais

Let Γ be a commutative with identity multiplicative hyperring and S ⊆ Γ be a multiplicatively closed subset of Γ. In this study, we will discuss the definition and general properties of weakly Γ-prime hyperideals. Let Ω be a hyperideal of Γ disjoint from S. We say that Ω is a weakly S− prime hyperideal of Γ if there exists an s ∈ S such that for all ϱ,σ ∈ Γ , if {0} ̸ = ϱ⋄σ ⊆ […]

Ver mais

How I built an AI-native Second Brain for engineering management — with Obsidian, Claude Code, and zero subscriptions. Seven unread Slack DMs. Three stacked 1:1s before lunch. A promotion packet review you can’t find. An incident follow-up thread someone archived. A decision from last quarter that lives in a Slack thread you can no longer search because it’s past the retention window. This is the job now. Not the work — the scaffolding around the work. Your brain […]

Ver mais

ARC-AGI has introduced a third stage of its famous benchmark. You can review it here. ARC-AGI3 distances itself from 1 and 2, developing towards a more genuine test of task acquisition. If you play demos of ARC-AGI3, you will see that they are beginning to mimic traditional environments seen in Reinforcement Learning research. Design Philosophy Easy for Humans, Hard for AI At the core of ARC-AGI benchmark design is the the principle of “Easy for Humans, Hard for […]

Ver mais

Andrej Karpathy is a name many people in the AI world will recognize. He was a professor at Stanford for a while, later became the head of Tesla’s AI division, worked at OpenAI, and, in my opinion, he also produces some of the best AI educational videos available today. His latest project, microGPT, however, is something I can only describe as a work of art. microGPT is a complete GPT implementation written in only about 200 lines of […]

Ver mais

Recently, differentiable causal discovery has emerged as a promising approach to improve the accuracy and efficiency of existing methods. However, when applied to high-dimensional data or data with latent confounders, these methods, often based on off-the-shelf continuous optimization algorithms, struggle with the vast search space, the complexity of the objective function, and the nontrivial nature of graph-theoretical constraints. As a result, there has been a surge of interest in leveraging super-structures to guide the optimization process. Nonetheless, learning […]

Ver mais