Training & Fine-tuning
A
Showing 61–85 of 85
-
Beyond Safe Answers: Segment-Aware Listwise Alignment for Reasoning Safety in Large Reasoning Models
-
Authorship attribution and aesthetic evaluation of AI poetry: a case study with Haiku
-
How Fyxer built an AI executive assistant people trustOpenAI showcases Fyxer's AI EA: 53% of drafts sent unedited, 90% retentionOpenAI profiles Fyxer, a UK startup whose AI executive assistant is trained on 500,000+ hours of human EA workflows. Email handling is split across 30–50 specialized fine-tuned models, and user edits feed a DPO self-training loop. 53% of drafts are accepted as written, 90-day retention tops 90%, and ARR grew from $1M to $32M in 2025.
-
RSIAgent: Autonomous Exploration for Recursive Self-improvement in New Environments
-
Parameter-Efficient Adaptation of Pretrained Language Models for Time-Series Forecasting
-
Reducing the Output-Mode Gap in Speech Language Models via Joint-Output On-Policy Distillation
-
EuroBirdPortal – Live bird movements across EuropeEuroBirdPortal maps live weekly bird movements across EuropeEuroBirdPortal aggregates bird observation records from across Europe and animates weekly changes in species counts and distribution on an interactive map. It offers a double-map view for comparing two species and a 52-week timeline playback, making seasonal migration patterns easy to follow. The citizen-science visualization drew attention on Hacker News.
-
TailscaleのVPNにAIエージェントを組み込める「Aperture」正式リリース。AIによるTailscaleやノードの操作も可能にTailscale ships Aperture AI gateway GA, adds MCPs for agent-driven network opsTailscale announced general availability of Aperture, an AI gateway that lets any node in a tailnet reach AI services without distributing API keys. It now bundles cost controls, guardrails, audit logs, MCP/API proxies, and in-gateway token purchasing. New Tailscale MCP and Tailscale SSH MCP let AI agents add nodes, SSH into machines, and deploy services via Aperture's chat UI, with existing ACLs enforced, human approval required for new nodes, and every action logged.
-
So you want to use OpenRouter?OpenRouter's auto-routing can vary model behavior by providerSimon Willison flags Mohamed Moustafa's caveats on OpenRouter: its single endpoint auto-routes to the cheapest backend, but providers differ in serving software and settings, so the same model can behave differently. provider.only pins routing.
-
Expert-Space Exploration in MoE Reinforcement Learning
-
Dynin-Robotics: Omnimodal Unified Diffusion Vision-Language-Action Model
-
Generative Retrieval for Unsupervised Text-Based Person Search
-
英国王立協会特集号に見る、世界モデルの最前線とAIの未来Royal Society issue maps world models; Sakana AI's David Ha co-authorsThe Royal Society's Philosophical Transactions A published a special issue, "World Models in Natural and Artificial Intelligence," co-authored in its opening article by Sakana AI CEO David Ha. Contributors argue scaling compute alone will not close the gap between what large models can do and what they understand, and tie world models to artificial life.
-
Parameter-Efficient Retrievers for Polish and European Languages
-
UniPart: Towards Zero-shot Language-Grounded 3D Part Segmentation for Embodied Interaction
-
Behavior Quotient Learning for Low-Rank Adaptation of LLM Agents
-
Evaluating Context Segmentation in Locally Deployable SLMs for Cybersecurity CTF Tasks
-
Scaling Clinical Judgment to Evaluate Medical AI
-
RunningTensor: Generalizing Linear Attention to Higher-Order Recurrent States
-
Cognition on Graph: Navigating Massive Knowledge Space via Cognitive Cycles and Bidirectional Graph-Text Synergy
-
SCQ: Stabilizing Conservative Q-Learning with Sigmoid-Bounded Entropy
-
Residual Vector-based Reconstruction as Long-Context Recall Regardless of Context Window Size
-
Doc2FRC: Length-Consistent Document-Level Machine Translation via Fixed-Range Chunking
-
Earth-Agent-Pro: Towards Real-World Full-Chain Earth Observation with Agents
-
フィジカルAI向け「データ収集工場」に"潜入" 実機写真で学習データ生成の流れを解説J-HRTI opens a physical-AI robot data factory in Narashino, JapanA five-company cross-industry consortium, J-HRTI, has started its Kanto Data Factory in Narashino, Chiba, producing training data for physical AI and robotics. Member firms share the collected data; ITmedia details the capture process with on-site photos.