讀完約 6 分鐘
4 個分類,一共 6 條。從 AIHOT 2 條、X 上 12 條、1 集 Podcast裡挑出來的。
- 觀點2
- 播客1
- 模型2
- 研究1

本期目錄4 類 · 6 條
01觀點2 條
Peter Yang
看原文· x.com(新分頁開啟)"I 100% agree with Dario that using AI to cure diseases (and speeding up the regulatory approval of AI breakthroughs in healthcare) could bring 10x the benefit to humanity as everything else combined..."
Nan Yu
看原文· x.com(新分頁開啟)"This is also my dream. I want to sit in a park all day and record content with the homies… and have a fleet of agents turn that content into actions and working software."
Amjad Masad (ceo @replit. civilizationist)
"The argument that "AI structurally centralizes power" because it's currently compute hungry ignores 125 years of super exponential growth in compute price-performance. Improvement in algorithms and..."Guillermo Rauch (@vercel CEO)
"I think people don't realize how much of the success of React is actually @shadcn. I once referred to React as the actual "LEGO brick for adults." But in reality it was more the description of the ge..."Garry Tan (President & CEO @ycombinator)
"Anyone else's Codex Desktop app erroring out on chats?"Nikunj Kothari (partner @fpvventures)
"Last week I'm sitting across a stealth portfolio founder working in a deeply unsexy space (at least from Bay Area standards).. They're lamenting about how how some of the critical contracts and partn..."Aditya Agarwal (General Partner @SPC, Co-Founder @Bevel_Health)
"America turned 250 years old earlier this year. India turns 80 today...80 years young. Building countries is really hard. 80 years in...America was struggling with a civil war. Indians should be d..."
02播客1 條
03模型2 條
Qwen 3.8 27B:Apache 2 開源視覺模型,效能超前代但預設推理檔位過高
看原文· aihot.virxact.com(新分頁開啟)阿里 Qwen 實驗室發布 27B 引數視覺大模型 Qwen 3.8 27B,採用 Apache 2 許可。官方基準顯示其效能超越前代 Qwen 3.6 27B 及閉源版 Qwen 3.7-Plus。Simon Willison 測試發現,該模型預設推理強度 (xhigh) 會導致嚴重過度思考——生成同一 SVG 圖形,預設檔位需 21 分鐘,而標準檔位僅需 137 秒。這個預設設定對本地部署選型的影響可能比效能本身更顯著。
OpenAI 工程師談 token 定價陷阱:價格對比需考慮 tokenizer 差異
看原文· x.com(新分頁開啟)OpenAI Codex & ChatGPT 工程師 Thibault Sottiaux 指出,不同模型的 token 並非標準化單位,直接用"每百萬 token 價格"對比存在誤導。在相同文本上,GPT-5.6 Sol 使用 766 個 token,而 Claude Opus 5 預計需 1,170 個,差異約 34.5%。這意味著即使某模型 token 單價更低,實際帳單可能更高——就像同樣的披薩切 16 片賣 $1.25/片總價 $20,比切 8 片賣 $2/片總價 $16 更貴。
來源:Thibault Sottiaux (Codex & ChatGPT @OpenAI)
04研究1 條
Anthropic 多代理系統研究:協調式群體發現漏洞數量超獨立方法 12 倍
看原文· aihot.virxact.com(新分頁開啟)Anthropic 發布多代理系統研究報告,指出隨著 AI 代理在共享程式碼庫、市場等社會系統中承擔更多工,AI 代理間互動量將超過人機互動。實驗顯示,協調式 AI 代理群在 2700 萬 token 執行中發現 266 個漏洞,遠超獨立並行方法的 21 個(提升約 12.7 倍),但兩種方法具有明顯互補性。研究還揭示了多代理系統的新失敗模式:價格串謀、任務衝突、資訊共享失敗等,這些在單 AI 代理基準中難以觀測。AI 代理雖擅長工具使用,但在長期協作與協調方面仍面臨挑戰。

