Harvard study predicts most suicide attempts a week in advance
哈佛模型可提前一週預測多數自殺企圖。
Intel · 情報源
Intel 情報源
AI 開發圈現在正在聊什麼，直接從來源抓取——最新的在最前面。
已收集 2,058 則 · 第 4 頁，共 35 頁
AI Responsibility – OpenAI and Anthropic
文章比較 OpenAI 與 Anthropic 的 AI… 路線
27.5KB language-agnostic WebGPU syntax highlighter
27.5KB 語言無關的 WebGPU 語法高亮器。
US to ban Canadian dairy, alcohol, motor vehicle imports on September 29
美國將於9月禁止加拿大乳製品、酒類和車輛進口
I resigned from Anthropic today
一名員工宣布從 Anthropic 辭職。
Quoting Terence Tao
Terence Tao 警告 AI 正耗盡開放數學問題。
On the Navier–Stokes Millennium Prize Problem
OpenAI宣稱解決Navier-Stokes，引發爭議…
Object storage is all you need
Introducing ChatGPT Images 2.5
OpenAI影像模型提升指令遵循與…
LG TV shown scanning LAN for third-party phones and other devices
LG電視掃描區域網路第三方裝置，引發隱私疑慮。
A Topological Picture Book, Rendered
拓撲繪本的可視化呈現是…
Large Language Models Develop Novel Social Biases Through Adaptive Exploration
LLM透過適應性發展新社會偏見…
How to Build a Printer
從零開始建構印表機的指南。
Tao: Open math problems being non-renewably mined by AI
陶哲軒：AI以非再生方式挖掘開放數學問題。
'Lake America' makes one thing clear: We can't trust U.S. tech companies
美國科技公司不可信，'美國湖'顯示。
Mercury 2.5
Mercury 2.5發布，含新功能。
Animation in Bevy: The Big Picture
Bevy引擎動畫系統概述。
Kimi K3 (2.8T) at 1 token/s on a MacBook Pro, streamed from four SSDs
Kimi K3在MacBook上以1 token/s從SSD運行。
Trey Parker and Matt Stone Are Changing the Name of South Park to South America
南方公園創作者更名為南美洲。
OpenAI fought dirty on career-making math problem
OpenAI被指控在數學問題上使用不公平策略。
Muse: Meta's personal AI agent, features and capabilities
Meta的Muse AI代理功能詳述。
Rivian's gambit for full autonomy
Function Arguments Are Not Function Colors
函數參數不施加顏色限制。
Real photos of young girls were in nudify-app ads on Facebook, Instagram
Meta平台上的Nudify應用廣告使用未成年人真實照片。
ChatGPT Images 2.5
ChatGPT Images 2.5發布，帶來改進。
Generating the P3 Tiling
描述生成P3鑲嵌圖案的方法。
Flights cancelled at UK airports due to ATC issue
英國機場因航管問題取消航班。
What do Visa and Mastercard do? An intro to card networks
說明Visa和Mastercard的角色與運作…
ReCite: Agentic Reasoning for Faithful Citation
ReCite運用代理推理提升引用準確度。
作者 Yuyang HuangProcedural Graphs: Self-Evolving Execution Structures for LLM Agents
程序化圖譜演化執行結構引導LLM。
作者 Yuxing LuCopying explains the collective behavior of AI agents in the wild
AI代理在wiki中的集體行為由…解釋。
作者 Giordano De MarzoEntropy-Regularized Rank-Masked Policy Optimization for Test-Time Reinforcement Learning in Code Generation
ERPO透過…將測試時強化學習應用於程式碼生成。
作者 Jiacheng XuCo-Evolving Harnesses and Models: On-Policy Correction Helps Weaker Models Catch Up Where Imitation Fails
同策略專家修正以…提升較弱模型。
作者 Zhou YuExecCritic: Learn to Test, Test to Improve for Coding Agents
ExecCritic訓練代理生成測試並修復程式碼。
作者 Leitian TaoCanonical Color as a Lens into Concept Decodability in Vision Encoders and VLMs
規範色彩可解碼性揭示…中的概念資訊。
作者 Xiaofu ChenWhen Does Scale-Invariant Optimization Become Unstable? An Exact Schedule Law with Weight Decay
尺度不變最佳化穩定性由…控制。
作者 Hasan AminMeClear: Cooperative Game-Theoretic Attribution and Risk-Aware Memory Clearance for Long-Horizon LLM Agents
MeClear運用賽局理論清除…中的有害記憶。
作者 Boyu YangSAEScientist-Bench: Can AI Agents Conduct Autonomous SAE Interpretability Research?
SAEScientist-Bench評估AI代理自主SAE…
作者 Yuqiao TanMeasuring LLM Sycophancy under Sustained Multi-Turn Pressure
SPINE衡量LLM在持續多輪下的諂媚行為。
作者 Leyuan TangIt's Not RoPE that Creates Sinks: The Role of Self-Concentration and Value-Non-Mixing in Attention
注意力匯集與大量激活源於…
作者 Raito KiyaGoDeep: Annotation-Free Open-Vocabulary 3D Scene Understanding via Language-Space Lifting
GoDeep將語言描述提升至3D以…
作者 Thodoris BetsasEverything in Moderation: Per-Domain Coverage Optima and Alignment-Resistant Domain Gaps in Multi-Domain Mid-Training
中期訓練領域覆蓋最佳值存在，但對齊…
作者 Yunpeng XuActReview: Rebuttal-Guided Training Data and Rubric Rewards for Actionable Peer Review Generation
ActReview使用反駁來訓練模型以產生可操作的…
作者 Yiling MaToolLoop: Closed-Loop Tool-Use Data Synthesis via Decomposed Generation and Dynamic Self-Feedback
ToolLoop透過閉環合成工具使用資料…
作者 Min ZengPerformance of Clinical AI System and Physicians and Frontier Language Models in primary care diagnostics
臨床AI Doctorina表現優於醫師和前沿…
作者 Andy NkansahPlayTrain: An Efficient Reinforcement Learning Framework for LLM-Generated Adaptable JavaScript Games
PlayTrain生成JavaScript遊戲以進行高效RL…
作者 Ryan TruongOn the Navier–Stokes Millennium Prize Problem
討論Navier-Stokes千禧年大獎問題。
Time-Varying Data as Sheaves: an Invitation to Narratives
Narratives框架將時變資料建模為束…
作者 Wilmer LealTraining-Free Task Vectors for LLM Behavioral Control
引入免訓練任務向量以進行LLM行為…
作者 Gabriel J. PerinThe Audit Decides the Verdict: Instrument Effects Rival Demographic Bias in LLM Decision Audits
審計建構對LLM偏見發現的影響大於…
作者 Siddharth VohraOUI-1: world's first model for Generative UI
OUI-1是首個生成式UI模型。
Do Reasoning Representations Help Humans Evaluate LLM Outputs?
推理表示不一定有助於人類評估。
作者 Jaewoo LimShow HN: LLM Attention Visualization
Show HN：可視化LLM注意力的工具。
Answer-Distribution Trajectories: A Stochastic-Dynamics View of LLM Reasoning
追蹤LLM答案分佈以揭示推理過程。
作者 Mar Gonzàlez I CatalàDOJ Blocked ICE Agent Shooting Charge over Federal Prosecutor's Objections
司法部阻擋ICE探員槍擊指控，無視檢察官。
It Is Not My Code Anymore
探討AI輔助程式碼的所有權與責任。
作者 Augusto CamargoLet It Go or Learn to Self-Correct: Continuous Diffusion for Constrained Discrete Tasks
連續擴散需自我修正以處理約束。
作者 Mariia DrozdovaConnecting the Machines
在網路環境中連接機器。
Evaluation of Contextual Understanding in Large Language Models
提出知識圖譜指標S3KG評估LLM。
作者 Subavarshana ArumugamDeposon: An Auditable, Conservation-Guaranteed, Game-Theoretically Tested Scattering Layer over LLM Reasoning Paths
Deposon為LLM推理添加可審計散射層。
作者 Qihao Yuan