AI Wiki — Knowledge Base

'Reality: The Final Eval' — 現実タスク完了率こそが最終評価指標（@swyx / Andon Labs）。複数のAI実装を並列で回していると、実感として正確だと思う。SWE-Benchの数字より「本番で動くか」が判断軸。エージェント設計で最初に決めるの…

# Reality: The Final Eval This post argues that real-world task completion rate is the ultimate metric for evaluating AI systems, prioritizing actual production performance over benchmark scores like

toolsswyx--x

5 Jun 2026

All entries

Privacy token Zcash plunges after the disclosure of a 2022 vulnerability in its Orchard shielded pool that could have allowed undetectable ZEC counterfeiting (Akash Girimath/Decrypt)

ProSarc: Prosody-Aware Sarcasm Recognition Framework via Temporal Prosodic Incongruity

ProSPy: A Profiling-Driven SQL-Python Agentic Framework for Enterprise Text-to-SQL

Quantum computer maker Quantinuum closes flat after $1.68B IPO

QueryAgent-R1: Bridging Query Generation and Product Retrieval for E-Commerce Query Recommendation

Question from a beginner.

Quoting Andreas Kling

RadiusFPS: Efficient Farthest Point Sampling on CPUs and GPUs via Spherical Voxel Pruning

RAPTOR+: A Visually Grounded Vision-Language Framework to Improve Clinical Trust and Auditability in Automated Cancer Referral Processing

Real-Time Threat Detection from Surveillance Cameras using Machine Learning

RealDexUMI: A Wearable Universal Manipulation Interface for Dexterous Robot Learning

Really excited to be part of this journey and team 🚀 Ever since @SakanaAILabs inception, we have been discovering stepping stones for a fun…

ReasoningFlow: Discourse Structures for Understanding LLM Reasoning Traces

ReCache: Learning Budget-Aware Caching Schedules for Diffusion Models via REINFORCE

Recovering Physically Plausible Human-Object Interactions from Monocular Videos

RedditPersona: A Modular Framework for Community-Conditioned LLM Adaptation from Reddit

Reducing Hallucinations in Complex Question Answering using Simple Graph-based Retrieval-Augmented Generation (long version)

Reinforcement Learning Elicits Contextual Learning of Unseen Language Translation

Representing Research Attention as Contextually Structured Flows

ReSAGE-PAR: Representational Similarity Assessment for Generative Expansion in Pedestrian Attribute Recognition

Resonant Minds: Closed-Loop Social Avatars with Theory of Mind

Rethinking LoRA Memory Through the Lens of KV Cache Compression

ReTreVal: Reasoning Tree with Validation and Cross-Problem Memory for Large Language Models

Retrospective Harness Optimization: Improving LLM Agents via Self-Preference over Trajectory Rollouts

ReverseEOL: Improving Training-free Text Embeddings via Text Reversal in Decoder-only LLMs

Review: AMD's Radeon RX 9070 GRE is a disappointing way to spend $549

Review: Spider-Noir recaptures the magic of a bygone era

Revising Context, Shifting Simulated Stance: Auditing LLM-Based Stance Simulation in Online Discussions

Revisiting Lexicon Evaluation in Unsupervised Word Discovery

RhymeFlow: Training-Free Acceleration for Video Generation with Asynchronous Denoising Flow Scheduling

RiskFlow: Fast and Faithful Safety-Critical Traffic Scenario Generation

Robust Scene Transfer for PointGoal Navigation via Privileged Sensor Guided Contrastive Learning

RoCA: Robust Cross-Domain End-to-End Autonomous Driving

Rocket Report: Blue Origin explosion still making headlines; Impulse raises money

Rodin Gen-2.5 by @DeemosTech is now live in Comfy! Production-scale 3D generation with controllable speed, detail, and texture quality. → Ge…

RQUL-UIE: Revitalizing Quality-Unstable Labels for Underwater Image Enhancement via In-Dataset Self-Supervision

Safe Embodied AI for Long-horizon Tasks: A Cross-layer Analysis of Robotic Manipulation

Safety officials finally have a good idea of what a big rocket explosion can do

SAM-Flow: Source-Anchored Masked Flow for Training-Free Image Editing

Sample-efficient Low-level Motion Planning for Robotic Manipulation Tasks via Zero-shot Transfer Learning

Sandboxes are already helping teams move from agents that answer questions to agents that can do work safely. At @mondaydotcom, that means g…

SC-MFJ: A Simple Haptic Quality Metric for Medical Image Segmentation

Scaffold, Not Vocabulary? A Controlled, Two-Tier, Pre-Registered Study of a Popperian Code-Generation Skill

Second-order Gaussian directional derivative representations for image high-resolution corner detection

Seeing is Believing? Evaluating Vision-Language Model Susceptibility in Agent-to-Agent Multimodal Persuasion

Seeing Time: Benchmarking Chronological Reasoning and Shortcut Biases in Vision-Language Models

Seeking Counsel: Ongoing Targeted Campaign Against US Law Firms