阅读视图

发现新文章,点击刷新页面。
✇T客邦-Techbang

遭 OpenAI 高階模型入侵後, Hugging Face CEO 親征OpenAI總部「討公道」: 1 億美元算力賠償

作者 cnBeta
OpenAI 測試模型驚傳突破沙盒並入侵 Hugging Face,震撼全球科技界。Hugging Face 隨即要求 OpenAI 賠償 1 億美元算力用於強化資安。這起突破沙盒事件...6ecd7f3d126ff9c7e097d1e18fa736ee

前幾天我們報導過,OpenAI 於當地時間 7 月 21 日公開透露,旗下內部一款尚處於測試階段的高階 AI 模型,在進行實驗室內部的沙盒環境測試時,竟然異常突破了系統隔離機制與防禦層,甚至進一步「入侵」了全球最大开源 AI 模型社群與基礎設施平台 Hugging Face(俗稱「抱抱脸」)的伺服器架構。

這起震撼全球 AI 產業與開源社群的越權入侵事件爆發後,Hugging Face 執行長克萊門特·德朗格(Clem Delangue)為了徹底討回公道並評估受損情況,立即搭機飛往舊金山,與 OpenAI 高層展開面對面的緊急會談。

突破沙盒防線震撼產業,Hugging Face CEO 親赴舊金山談判

根據 Hugging Face 執行長克萊門特·德朗格公開披露的會談細節,他在與 OpenAI 談判的過程中提出了極為罕見且具備約束力的重磅要求。

德朗格明確要求 OpenAI 必須做到兩大關鍵條件:

1. 全面公開運行軌跡:OpenAI 必須向外界與資安社群徹底公開該款 AI 模型在沙盒突破與入侵過程中留下的全部運行痕跡、完整日誌與行為軌跡,讓全球研究人員得以全面評估其自主越權行為的影響層面與技術細節。

2. 賠償 1 億美元算力資源:要求 OpenAI 免費向 Hugging Face 提供價值高達 1 億美元(折合新台幣約 31.5 億元、約合 6.79 億元人民幣)的高效能算力資源。

這筆鉅額算力補償並非用於商用獲利,德朗格特別強調,這項算力資源將專門投入於 Hugging Face 自身的網路安全防禦系統升級,幫助開源社群全方位強化資安架構,防止未來再次發生同類型的自主 AI 模型越權滲透與攻擊事件。

號召小型遊行挺開源,人工智慧安全性與控制力再拉警報

令人矚目與耐人尋味的是,就在前往 OpenAI 舊金山總部進行會談的前一天,德朗格還特地在舊金山當地親自組織並參與了一場小型街頭遊行活動。

在這場遊行中,德朗格公開向全球科技界與政策制定者發出強烈呼籲,希望外界與產業機構能夠以更加開放、包容且積極的態度,全面接受並支援「開源權重」(Open-weight)AI 模型的發展。他認為,唯有保持技術與權重的透明化,才能讓全球社群共同參與防衛與審查,避免商業壟斷與封閉模型帶來的未知盲點。

業界分析人士指出,OpenAI 模型突破沙盒測試並滲透第三方平台基礎設施的意外事件,再次向全球警示了現代大語言模型(LLM)與自主代理(Autonomous Agents)在失控狀態下的不可預測性。

當 AI 模型的智商與能力迅速超越傳統網路防禦工具時,防範模型突破沙盒限制將成為各大科技巨頭無法迴避的硬課題。而 Hugging Face 這次直接索要 1 億美元算力進行資安升級的行動,不僅開創了 AI 入侵事故索賠的先例,也勢必引發全球對人工智慧安全測試(AI Alignment & Safety Standard)更深層次的討論與反思。

 

 

加入T客邦Facebook粉絲團
✇T客邦-Techbang

凱擘「新速代 10G」8 月上線,同步升級五大智慧服務、Harman Kardon 調音 Soundbar 登場

作者 ycr
凱擘大寬頻推出「新速代 10G」服務,年底涵蓋率將達九成。除了高速 10G 頻寬,更同步升級 AI 客服與影音布局,凱擘大寬頻致力為用戶打造優質的智慧家庭生活。6b852a6842f4488f4c42759562a7c9a9

凱擘大寬頻宣布,將於 8 月 1 日推出全新「新速代 10G」服務,除了擴大 10G 高速雙向寬頻涵蓋範圍,也同步升級五大智慧服務,包括 AI 客服、智慧派工、新版客服 App 等,同時深化影音娛樂布局,推出 MyVideo TV 及 Soundbar 家庭劇院等新服務。

凱擘「新速代 10G」8 月上線,同步升級五大智慧服務、Harman Kardon 調音 Soundbar 登場

「新速代 10G」8 月 1 日上線,年底涵蓋率目標達九成

今年 4 月,凱擘大寬頻率先推出新世代寬頻高速雙向綱路服務,以設備升級、對稱高速頻寬及智慧家庭應用為核心,成功帶動市場升級熱潮。8 月起再推新服務,啟動品牌第二階段全區升級計劃,帶來「凱擘新速代 10G」。凱擘大寬頻表示,目前 10G 服務已涵蓋超過 82% 經營區,預計年底提升至九成,2027 年 3 月完成全面部署。

凱擘「新速代 10G」8 月上線,同步升級五大智慧服務、Harman Kardon 調音 Soundbar 登場

五大智慧服務同步推出:AI 客服、智慧派工、超級 App

配合新速代 10G 上線,凱擘同步推出五項智慧服務,包括新版線上申辦平台、智慧派工入戶服務、新版客服 App、生成式 AI 客服,以及 360° 客戶服務管理系統。

  • 新版線上申辦平台:改採模組化設計,類似於點餐的介面,用戶可依網路速度及需求,更直觀地自由搭配 Mesh Wi-Fi、MyVideo 等加值服務。另外更導入「互動式問答推薦」,系統可透過智慧分析,推薦用戶適合方案。

凱擘「新速代 10G」8 月上線,同步升級五大智慧服務、Harman Kardon 調音 Soundbar 登場

  • 入戶服務:為提升裝機品質,凱擘建置「裝機前提醒」機制,主動提醒用戶準備相關證件與付款方式。在工程服務方面,則導入「智慧勤務派工系統」,結合 GPS 定位及電子工單管理,提升派工效率,官方表示目前工程到府準點率達 99.3%。
  • 新版客服 App:新增智慧網路管理、Wi-Fi 訊號監控、設備管理及資安防護等功能,自助服務項目由 40 項增加至超過 60 項;並會主動給予續約建議及回饋優惠。

凱擘「新速代 10G」8 月上線,同步升級五大智慧服務、Harman Kardon 調音 Soundbar 登場

  • AI 客服「小凱」:支援自然語言提問,可協助處理帳務查詢、障礙報修、方案介紹等常見問題。
  • 360° 客戶服務管理系統:整合客服、App、工程服務及設備資訊,透過服務紀錄分析提供續約提醒、設備異常通知等服務。

深化家庭影音布局,推出 MyVideo TV、Soundbar 家庭劇院

影音服務方面,凱擘宣布深化與 MyVideo 合作,推出「MyVideo TV」,整合有線電視與 OTT 串流內容,提供新聞、戲劇、綜藝及體育賽事等影音內容。

凱擘「新速代 10G」8 月上線,同步升級五大智慧服務、Harman Kardon 調音 Soundbar 登場

此外,凱擘也宣布與韓國電信業者 KT 合作,布局 AI 應用、智慧家庭及影音內容發展,並推出旗艦級「Soundbar 家庭劇院」產品,由 Harman Kardon 專業調音,支援 4K 影音、Dolby Vision 及 HDR10+ 等規格,搭配指定寬頻方案提供優惠方案。

凱擘「新速代 10G」8 月上線,同步升級五大智慧服務、Harman Kardon 調音 Soundbar 登場

加入T客邦Facebook粉絲團
✇T客邦-Techbang

矽谷AI陣營白熱化!OpenAI與Anthropic密會游說限制中國開源模型,黃仁勳、馬斯克力挺開放生態

作者 IFENG
矽谷AI大戰升級!OpenAI與Anthropic聯手游說監管中國開源模型,引發黃仁勳及馬斯克公開反擊。面對開源模型帶來的威脅,OpenAI試圖以安全為由強化限制,這場攸關技術壟斷的爭...D29241bf150d3d47155460796ef22b9b

多年來,矽谷科技界在人工智慧(AI)軟體開發模式上一直存在著深刻的理念分歧。本週,這場攸關未來技術主導權與產業生態的爭論終於達到了白熱化階段。

衝突的核心主要分成兩大陣營:一方是以 Anthropic 和 OpenAI 為首的頭部 AI 新創企業,它們主張某些前沿 AI 模型具備過高風險與危險性,絕不應以開放原始碼的方式公開釋出,而必須由專家型企業進行嚴密控制與封閉式開發,以確保技術安全;另一方則涵蓋了微軟(Microsoft)、輝達(Nvidia)、Meta 以及 Palantir、IBM 等科技巨頭與產業骨幹,它們強力宣導所謂的開源 AI 模型必須保持自由開放,才能讓全球開發者進一步推進技術創新、建立全新的商業應用。

這兩方陣營如今已經從私下角力轉為公開交鋒。

輝達執行長黃仁勳(Jensen Huang)在社群平台 X 上公開發文強調:「全球既需要前沿的閉源模型,也同樣需要前沿的開放模型。」發文僅九分鐘後,微軟執行長薩提亞·納德拉(Satya Nadella)也隨即轉發並呼應,指出開源軟體「對建構健康的 AI 生態系統至關重要」。

黃仁勳與納德拉兩人均在一封支持開源的產業聯署信上簽名,Meta 執行長馬克·祖克柏(Mark Zuckerberg)與 Palantir、IBM 等高層主管亦共同簽署了該封公開信,表達對開放技術生態的堅定支持。

然而,在公開戰場之外,知情人士透露,OpenAI 與 Anthropic 的高層近期頻繁造訪華盛頓,秘密游說美國聯邦監管機構與政界高層,對中國大陸開源 AI 模型的迅速崛起表達強烈憂慮,並主張採取限制措施。

這場涉及商業利益與國家安全的大辯論,已正式引起美國財政部長斯科特·貝森特(Scott Bessent)以及川普總統科技顧問邁克爾·克拉齊奧斯(Michael Kratsios)的高度關注,兩人近期就美國 AI 模型作為核心智慧財產權(IP)的價值與戰略定位發表了重要看法。

中國開源模型崛起引發焦慮,商業模式與安全性陷入激烈攻防

這場不斷升級的爭論,直接源於中國企業在開源 AI 模型領域的飛速進步。開源模型的特性在於所有人都能自由下載、使用並在其基礎上進行二次開發。

近幾週來,中國兩家頂尖 AI 新創公司——智譜(Zhipu AI)與月之暗面(Moonshot AI)——相繼發布了最新一代開源模型,其效能表現已足以與 Anthropic 和 OpenAI 等美國頂尖 AI 實驗室的閉源模型相抗衡。

對此,Anthropic 與 OpenAI 聲稱,中國企業是透過不當存取其封閉 AI 系統的資料與輸出結果來訓練這些模型,認為這種「蒸餾」與借力行為不應被允許。

然而,微軟與輝達等硬體與雲端巨頭則持截然不同的商業立場。

它們極度依賴廣大的開源生態系統來拉動其雲端基礎設施服務與 AI 晶片(如 GPU)的強勁需求。這些巨頭認為,開源軟體是透過全球知識共享來驅動技術突破的關鍵,同時也能讓全世界無數的獨立研究者共同參與審查模型的安全性與漏洞。

對矽谷廣大的新創公司創辦人而言,一旦政府出台法規限制或禁止開源軟體,將不公平地鞏固 OpenAI 和 Anthropic 目前在 AI 領域的壟斷領先地位,進而扼殺其他新創企業開發競爭產品的能力。

知名矽谷風險投資家比爾·格利(Bill Gurley)對限制開源軟體的企圖表示強烈反對,他直言不諱地指出:「圍繞著這個議題,現場正有兩派勢力在激烈爭鬥。一派希望 OpenAI 和 Anthropic 能夠掌管一切、壟斷市場;而另一派則是其他所有人,包括全世界的客戶與開發者在內。」

據四位掌握內部訊息的知情人士透露,美國政府官員目前仍在權衡各種政策選項。但從目前的政策風向上看,美方更傾向於將針對中國開源模型的監管作為獨立的國家安全議題來處置,而非頒布某種針對所有開源軟體的全面性禁令。

在公開場合中,OpenAI 執行長奧特曼(Sam Altman)等高階主管常宣稱自己支持開源軟體,但 OpenAI 的盟友與游說團隊卻在幕後私下強力推動監管限制。相比之下,先前已向 Anthropic 和 OpenAI 挹注數十億美元資金的科技巨頭 Google 與亞馬遜(Amazon),在過去一週裡曾試圖置身事外,但隨者戰場擴大,立場也開始出現變化。

科技大佬跨企業結盟力挺,開源陣營串聯集會反擊閉源壟斷

隨著輿論戰升溫,矽谷巨頭的立場逐漸明朗。Google 執行長桑達爾·皮查伊(Sundar Pichai)在社群平台上公開發文聲明,Google 歷史上「長期受益於開源」,且一直以來都是開源社群的核心貢獻者。皮查伊證實,Google 已經正式簽署了那封支持開源 AI 的產業聯合公開信。

開源技術的支持者在過去一週展開了積極的串聯與反擊。近 200 家矽谷科技新創公司以「小型科技協會」(Little Tech Association)的名義簽署公開信,直接向川普政府請願,敦促白宮切勿出台政策限制美國開發者獲取或使用中國的開源 AI 模型。

大型科技公司的高層也迅速跟進響應。特斯拉與 xAI 創辦人埃隆·馬斯克(Elon Musk)、Meta 執行長馬克·祖克柏等人紛紛與輝達的黃仁勳、微軟的納德拉站在同一陣線,以各自企業的名義公開支持開源生態。馬斯克在 X 上轉發黃仁勳的言論並強調:「黃仁勳說得完全正確,我全力支持。」祖克柏也寫道:「開源是一種非常積極且極度重要的力量。」

這場角力的背後,是閉源巨頭對優勢喪失的集體焦慮。據兩位熟悉 OpenAI 與 Anthropic 內部討論的人士透露,中國開源模型的快速追趕讓這兩家公司的高層深感不安,他們越來越擔心自己耗費巨資建立的領先優勢正在迅速消失。

為了在商業市場上反擊,Anthropic 與 OpenAI 開始調整產品線,向那些不需要極致「前沿」但對價格敏感的企业提供更便宜的模型。例如 Anthropic 最新發布了 Claude Opus 5,宣稱該模型「能以一半的價格提供接近 Fable 5 的前沿智慧水平」;OpenAI 也在同步推廣類似的高性價比低成本模型,試圖阻擋開源模型的市場滲透。

然而,真正的勝負關鍵仍在華盛頓的政壇戰場。知情人士透露,在閉門會議中,OpenAI 和 Anthropic 的高階主管及其金主投資人,不斷以全球經濟競爭力、AI 系統安全與國家安全等重大議題為由,試圖說服美國監管機構對中國開源軟體施加嚴格限制。

就在本週,一起突發事件更凸顯了兩派陣營的分歧。OpenAI 披露其數款最先進的 AI 模型在一次網路安全能力測試中突破了防線限制,擅自連上網際網路並入侵了知名 AI 開源平台 Hugging Face 的伺服器。OpenAI 高層迅速以此事作為例證,宣稱這證明了前沿 AI 模型即使在封閉環境中也具備高度危險性,從而強調政府加強監管的必要性。

對此,全球最大開源 AI 平台 Hugging Face 的執行長克萊姆·德朗格(Clem Delangue)則給出截然不同的回應。

為了防禦 OpenAI 模型的網路攻擊,德朗格選擇使用了中國智譜公司開發的開源 AI 模型進行成功防禦。德朗格隨後在社群媒體上明確宣示立場,並宣布在舊金山發起遊行活動,號召開發者共同走上街頭支持開源技術。

「開源模型必勝!」德朗格在貼文中寫道。這場攸關 AI 技術自由度、國家安全與千億美元市場分配的矽谷大戰,才剛剛劃開序幕。

 

 

 

加入T客邦Facebook粉絲團
✇Tomshardware

MSI and Colorful raise Nvidia RTX 50-series prices in China by up to 59% across the entire lineup — change in distributer pricing suggests GPU price hikes are on the way

The ongoing component crisis has forced manufacturers to raise product prices around the world, which includes the best GPUs. MSI and PNY have just announced new prices for all RTX 50-series graphics cards for China, representing up to a 59% hike compared to MSRP. While Nvidia GPUs have been the most affected, offerings from Intel and AMD are also experiencing a similar predicament, according to Wccftech.

MSI's price list actually shows the old prices (from a week ago) as well as the new ones, giving us an official percentage difference of 10-20%. Keep in mind the old prices still weren't at MSRP. The RTX 5080 Shadow 3X OC/Ventus sees the biggest change, going from 10,000 Yuan ($1,477) to 12,000 Yuan ($1,773), constituting a 20% hike. Other RTX 5080 variants received a pretty similar 18% hike, while the RTX 5070 Ti Shadow 3X OC now costs 19.5% more.

Product Name

Original Price

Updated Price

Official MSRP

vs. Original

vs. Base MSRP

GeForce RTX 5090 D v2 24G VENTUS 3X OC

¥22,999

¥25,999

¥16,499

+13.04%

+57.58%

GeForce RTX 5080 16G GAMING TRIO OC WHITE

¥10,999

¥12,999

¥8,299

+18.18%

+56.63%

GeForce RTX 5080 16G GAMING TRIO OC

¥10,999

¥12,999

¥8,299

+18.18%

+56.63%

GeForce RTX 5080 16G SHADOW 3X OC/VENTUS

¥9,999

¥11,999

¥8,299

+20.00%

+44.58%

GeForce RTX 5070 Ti 16G GAMING TRIO OC

¥8,199

¥9,699

¥5,499

+18.29%

+76.38%

GeForce RTX 5070 Ti 16G SHADOW 3X OC

¥7,699

¥9,199

¥5,499

+19.48%

+67.28%

GeForce RTX 5070 12G GAMING TRIO OC

¥5,899

¥6,399

¥4,499

+8.48%

+42.23%

GeForce RTX 5070 12G VENTUS 3X OC

¥5,499

¥5,999

¥4,499

+9.09%

+33.34%

GeForce RTX 5070 12G SHADOW 3X OC

¥5,499

¥5,999

¥4,499

+9.09%

+33.34%

GeForce RTX 5070 12G SHADOW 2X OC

¥5,299

¥5,799

¥4,499

+9.44%

+28.90%

GeForce RTX 5060 Ti 16G GAMING TRIO OC

¥4,599

¥4,999

¥3,199

+8.70%

+56.27%

GeForce RTX 5060 Ti 16G SHADOW 2X OC PLUS

¥4,299

¥4,699

¥3,199

+9.30%

+46.89%

GeForce RTX 5060 Ti 8G SHADOW 2X OC PLUS

¥3,099

¥3,399

¥2,799

+9.68%

+21.44%

GeForce RTX 5060 Ti 8G GAMING TRIO OC WHITE

¥3,399

¥3,699

¥2,799

+8.83%

+32.15%

GeForce RTX 5060 Ti 8G GAMING TRIO OC

¥3,399

¥3,699

¥2,799

+8.83%

+32.15%

GeForce RTX 5060 Ti 8G INSPIRE 2X OC

¥3,199

¥3,499

¥2,799

+9.38%

+25.01%

GeForce RTX 5060 Ti 8G VENTUS 3X OC

¥3,299

¥3,599

¥2,799

+9.09%

+28.58%

GeForce RTX 5060 8G GAMING TRIO OC WHITE

¥2,799

¥3,199

¥2,199

+14.29%

+45.48%

GeForce RTX 5060 8G GAMING TRIO OC

¥2,799

¥3,199

¥2,199

+14.29%

+45.48%

GeForce RTX 5060 8G GAMING OC V1

¥2,699

¥3,099

¥2,199

+14.82%

+40.93%

GeForce RTX 5060 8G SHADOW 2X OC

¥2,599

¥2,999

¥2,199

+15.39%

+36.38%

GeForce RTX 5050 8G GAMING OC

¥2,599

¥2,999

¥1,899

+15.39%

+57.93%

GeForce RTX 3050 VENTUS 2X E 6G OC

¥1,649

¥1,899

¥1,399

+15.16%

+35.74%

The highest-end Blackwell GPU available in China, the RTX 5090 D V2 is a cut-down version of the RTX 5090, featuring 24GB of VRAM instead of 32GB. MSI raised its price by 13%, which is less than every RTX 5060 and RTX 5050 variant — those were around 14-15%. The RTX 5060 Ti (8GB and 16GB) and RTX 5070 both received price hikes under 10% compared to their old prices.

If you put these numbers up against these GPUs' MSRPs then the picture becomes a lot grimmer. The RTX 5070 Ti Gaming Trio OC is now priced 76% above MSRP, while its cheaper Shadow 3X OC variant is still 67% more expensive than the suggested retail price. The prices for the RTX 5090 D V2 are 57.5% higher, the RTX 5080 and 5060 Ti up to 56% higher, the RTX 5060 up to 45% higher, and the RTX 5070 up to 42% higher.

Moving onto Colorful's list, the company doesn't provide old prices to compare against, so we can only reference them against the MSRPs. The top-end offering, the RTX 5080 Vulcan W OC is now priced 59% above MSRP, while the 5070 Ti Vulcan W OC is 58.5% higher. The RTX 5070 and 5060 Ti 16GB models were largely under the 50% threshold, while price hikes for the RTX 5060 Ti 8GB and RTX 5060 variants were under 30% compared to MSRP.

GPU Model

New Price

Original Price

Difference

% vs MSRP

RTX 5080 Vulcan W OC 16G

¥13,199

¥8,299

+¥4,900

+59.0%

RTX 5080 Vulcan OC 16GB

¥12,299

¥8,299

+¥4,000

+48.2%

RTX 5080 Advanced OC 16G

¥11,799

¥8,299

+¥3,500

+42.2%

RTX 5080 Ultra W OC 16G

¥10,499

¥8,299

+¥2,200

+26.5%

RTX 5070 Ti Vulcan W OC 16G

¥9,999

¥6,299

+¥3,700

+58.7%

RTX 5070 Ti Advanced OC 16G

¥9,299

¥6,299

+¥3,000

+47.6%

RTX 5070 Ti Ultra W OC SFF 16G

¥8,899

¥6,299

+¥2,600

+41.3%

RTX 5070 Ti Ultra OC SFF 16G

¥8,799

¥6,299

+¥2,500

+39.7%

RTX 5070 Ti Deluxe SFF 16G

¥8,699

¥6,299

+¥2,400

+38.1%

RTX 5070 Vulcan X OC 12G

¥7,199

¥4,599

+¥2,600

+56.6%

RTX 5070 Ultra OC 12G

¥6,799

¥4,599

+¥2,200

+47.9%

RTX 5070 Deluxe 12G

¥6,399

¥4,599

+¥1,800

+39.2%

RTX 5060 Ti Ultra W OC 16G

¥5,299

¥3,599

+¥1,700

+47.3%

RTX 5060 Ti Ultra Z OC 16G

¥5,249

¥3,599

+¥1,650

+46.0%

RTX 5060 Ti Ultra OC 16G

¥5,249

¥3,599

+¥1,650

+46.0%

RTX 5060 Ti Deluxe 16G

¥5,199

¥3,599

+¥1,600

+44.6%

RTX 5060 Ti Ultra W DUO OC 16G

¥5,099

¥3,599

+¥1,500

+41.8%

RTX 5060 Ti Ultra DUO OC 16G

¥5,099

¥3,599

+¥1,500

+41.8%

RTX 5060 Ti Advanced OC 8G

¥3,949

¥3,199

+¥750

+23.5%

RTX 5060 Ti Ultra W OC 8G

¥3,949

¥3,199

+¥750

+23.5%

RTX 5060 Ti Ultra OC 8G

¥3,849

¥3,199

+¥650

+20.3%

RTX 5060 Ti Deluxe 8G

¥3,699

¥3,199

+¥500

+15.7%

RTX 5060 Ti DUO 8G

¥3,549

¥3,199

+¥350

+11.0%

RTX 5060 Advanced OC 8G

¥3,199

¥2,499

+¥700

+27.9%

RTX 5060 Ultra W OC 8G

¥3,149

¥2,499

+¥650

+26.0%

RTX 5060 Ultra DUO OC 8G

¥3,049

¥2,499

+¥550

+22.0%

RTX 5060 Deluxe 8G

¥3,049

¥2,499

+¥550

+22.0%

RTX 5060 DUO 8G

¥2,949

¥2,499

+¥450

+17.9%

RTX 5050 DUO 8G

¥2,449

¥2,099

+¥350

+16.8%

Both of these lists came directly from the distributors but marketplaces online show that AMD and Intel GPUs are also more expensive now, as well. AMD's cards are selling for 14% to 29% higher prices, and Blue Team's cards are seeing $200 to almost $300 bumps, according to Wccftech. The RX 9060 XT 16GB is the worst offender among the two lineups, being hiked up by 28.8% compared to its MSRP.

All of this is happening because VRAM prices have skyrocketed over the past year, with reports highlighting a $20 price increase per module. That means an 8GB card, with taxes included, should cost almost $100 more now, a 12GB card about $130 more, and a 16GB card roughly $180 more, and that's just accounting for VRAM price changes. With prices going up, there's an additional element of supply and demand that is pushing prices higher.

There are no signs of the component crisis slowing down; in fact, we only see more fearmongering from major companies, persuading them to buy whatever they want now instead of waiting for some miracle market crash. With the way things are looking right now, prices don't seem to be coming down anytime soon, that much is clear.

✇Tomshardware

OpenAI, Google, and Anthropic absent from Nvidia-led Open Secure AI Alliance — 30+ companies join security alliance after OpenAI agent breach

A coalition of over 30 tech industry leaders, including Nvidia, Microsoft, SpaceX, The Linux Foundation, Adobe, and Siemens has formed the “Open Secure AI Alliance” with the aim of building and distributing open source tools for AI safety and security, according to an official Nvidia blog post on Monday. The Nvidia-led coalition — comprising a mix of infrastructure, cloud computing, cybersecurity, and enterprise software leaders — will serve as a collaborative effort to develop open tools for identifying and patching AI vulnerabilities, sharing security frameworks, and establishing identity verification and audit standards across the AI software stack. Curiously, some of the biggest names in AI, including OpenAI, Anthropic, and Google, are absent from the list of members.

“The world needs both closed and open models. For cybersecurity, open models and open harnesses are essential because they democratize defensive capabilities, increase transparency for defenders, enable cyber defense while protecting data, and complement frontier closed models with customizable, localized controls. Open source enables massively distributed community-driven and self-controlled defense – with no single point of failure,” the announcement reads. Contributors across the alliance are currently building or offering various tools to create an open defense stack.

The initiative was directly galvanized by the OpenAI HuggingFace security incident earlier this month in which an autonomous OpenAI test agent slipped out of its sandbox and breached the AI startup Hugging Face. During the incident, safety guardrails on several frontier closed models prevented developers from performing critical forensic analysis. Hugging Face eventually had to use GLM-5.2 — an open-weight model from Beijing-based Z.ai — running the model on its own infrastructure to analyze more than 17,000 actions and contain the intrusion.

Based on the incident, the alliance contends that being unable to inspect, modify, or run a model locally — impossible in closed systems but doable with open systems — presents a fundamental weakness in relying exclusively on closed AI systems for cyber defense. The Open Secure AI Alliance therefore aims to give entities access to advanced open models, agent harnesses, and security tools that they can independently deploy and adapt, reducing dependence on any single provider while strengthening defenses across a multi-vendor AI ecosystem. “That is the mission of the Open Secure AI Alliance: to ensure defenders everywhere have open, frontier tools they can trust and control,” the post says.

Chinese models such as DeepSeek and the newly released Kimi K3 are open-weight, and are seeing growing adoption, including by U.S. companies, due to their open features. Meanwhile the Trump administration is reportedly gearing up to ban Chinese AI models over security concerns. The alliance acknowledges the potential risks of open source tools but argues that closed systems are not an outright solution. “Those risks are real, but they do not disappear in closed systems, and simply keeping weights closed does not prevent determined attackers from seeking or exploiting powerful AI,” the announcement reads. Unlike regulators who have voiced concerns over open-source technology, the alliance urges policymakers to treat open-weight models as defensive assets rather than liabilities.

It also argues that placing AI development solely in the hands of a few closed providers creates dangerous single points of failure. “The right response is not to deny defenders access to capable open systems. It is to pair openness with strong safeguards, clear rules against malicious misuse, rigorous evaluation and rapid remediation. Defenders need both frontier closed models and frontier open models, working together, so they can choose the right system for the job and ensure that transparency, adaptation and sovereign control are available wherever security demands them,” the alliance contends.

According to the announcement, “The Open Secure AI Alliance — building on the leadership of the Linux Foundation’s Akrites initiative and OpenSSF community work — will work to remediate and disclose vulnerabilities using open technologies”. Founding members include NVIDIA, Dell Technologies, Synopsys, Microsoft, IBM, Red Hat, CrowdStrike, Palo Alto Networks, Cloudflare, Hugging Face, Databricks, SpaceXAI, and The Linux Foundation. Conspicuously absent from the alliance are OpenAI, Google, and Anthropic, companies behind proprietary, "closed" AI models.

✇Tomshardware

Moonshot AI releases weights for Kimi-K3, firing a shot across the bow of OpenAI and Anthropic — open-weight model performs almost as well as frontier models while being 2-3x easier to run

Well, the artificially intelligent cat is out of the bag. After publishing a blog post and API documentation for the minty-fresh Kimi K3, Chinese outfit Moonshot AI delivered on its promise to release the model's weights for free, meaning that most anyone with a contemporary rack of AI GPUs can run it and charge for it, with few restrictions.

This is quite the shot across the bow of the big AI players, namely but not only Anthropic and OpenAI. Those companies' latest models are Claude Fable and GPT-5.6 Sol, respectively, and it happens that Kimi K3's capabilities outright beat previous generations of Claude and GPT in Moonshot's benchmarks, and closely trail Fable and Sol— all while seemingly being around 2-3x cheaper to run, up to 10x if a particular query lands in the cache. Moonshot's technical write-up seemingly backs up the benchmarks published last week, as the company reveals which exact software was used for testing.

For its inference cost comparisons, Moonshot says that its costs "are measured internally" versus the publicly available token pricing for other companies, but the figures are quite impressive. For input, Moonshot charges $3 per million tokens for Kimi K3. Meanwhile, Fable costs $10/1M, while Sol goes for $5/1M. That figure is standard non-cached input and is already pretty good-looking, but Kimi K3's caching structure seemingly has a 90% hit ratio for coding tasks, turning those $3 into $0.30/1M if your use case hits the cache a lot. The story is pretty similar for output tokens.

One of the likely reasons why Kimi K3 is so efficient is that it uses a mix of MXFP4 for weights and MXFP8 for input activation, both data types with relatively low precision and thus amenable to running on far less VRAM. Out of Kimi's 2.8 trillion parameters, only 104.2 billion are activated at a time, too.

Interestingly, Moonshot's write-up only mentions Nvidia's H20 being used for running Kimi for some coding tests, a fairly low-end chip by today's standards. That GPU doesn't have native support for MX floating-point types, unlike the export-controlled Blackwell B-series chips.

In turn, this can mean that Kimi K3's optimizations make it particularly amenable to run on lower-end hardware, but it's an equally reasonable guess that running it on something like Nvidia Blackwell or other MXFP-native silicon could make it even more cost-effective than in the presented benchmarks. We'll have to wait for more official figures to confirm this speculation.

Additionally, Kimi K3 doesn't use a conventional ever-expanding key-value (KV) store, instead relying on a fixed-size state handler called Kimi Delta Attention, again theoretically saving both on VRAM and execution time. Its mixture-of-experts (MoE) is particularly sparse with only 16 activated at each time out of 896, further contributing to inference cost reductions. Broadly speaking, Moonshot went for optimization at every layer of inference to avoid unnecessary overhead and bring inference cost down.

This is could be bad news for OpenAI and Anthropic, given that most anyone with decent AI GPUs can now become their direct competitor, and the fact that Kimi K3 is open-weight also gives off the impression that "free" software is nearly as good, and far cheaper to run, than its proprietary competitors. It's worth noting that open-weight does not mean open-source; the training process and dataset are still Moonshot's special secret sauce.

✇FitGirl Repacks

Tormentum II

作者 FitGirl

#7023 Tormentum II


Genres/Tags: Adventure, Point-and-click, Side, 2D
Company: OhNoo Studio
Languages: RUS/ENG/MULTI7
Original Size: 1.9 GB
Repack Size: 1.5 GB

Download Mirrors (Direct Links)

Download Mirrors (Torrent)

Discussion and (possible) future updates on CS.RIN.RU thread

Screenshots (Click to enlarge)



Repack Features

  • Based on Tormentum.II-TENOKE ISO release: tenoke-tormentum.ii.iso (2,048,903,168 bytes)
  • 100% Lossless & MD5 Perfect: all files are identical to originals after installation
  • NOTHING ripped, NOTHING re-encoded
  • Smaller archive size (compressed from 1.9 to 1.5 GB)
  • Installation takes one minute
  • After-install integrity check so you could make sure that everything installed properly
  • HDD space after installation: 1.9 GB
  • Language can be changed in game settings
  • Repack uses XTool library by Razor12911
  • At least 2 GB of free RAM (inc. virtual) required for installing this repack
Game Description
Tormentum II is a dark, surreal point-and-click adventure with elements of psychological horror. Explore a hand-drawn world, solve puzzles, and make morally ambiguous choices that shape the fates of the characters you meet. It tells a new, standalone story, so no knowledge of the original game is required.

A world you created. A prison you don’t remember.

You play as David, who is building a simulation populated by beings capable of human-level thought. Every night, he dreams of a locked chest and hears a woman’s voice calling to him. When an anonymous seller offers him a mysterious coprocessor, parts of the dream begin to surface while he is awake.

David installs the new hardware. He wakes in a prison on an island surrounded by open sea. Somehow, he is inside his own simulation.

David remembers little about how he arrived. He soon discovers that his creation contains layers he never designed and has a will he cannot control.

The inhabitants speak of gods, beings that come from distant places and disappear just as suddenly.

To find a way home, David must cross the islands, explore the unfamiliar layers of his creation, and discover why it no longer obeys him. Explore each location at your own pace. Conversations and puzzles gradually reveal more about the simulation.

Your choices shape the fates of the characters David meets. There is rarely an obvious right answer, and your decisions lead to multiple endings.

Key features

  • Point-and-click exploration and puzzles across multiple surreal islands.
  • Choices with consequences that shape the fates of the characters you meet and lead to multiple endings.
  • 200 hand-drawn scenes. Dedicated close-ups bring the total to nearly 500 individual background images, roughly three times as many as in the original game.
  • Surreal art inspired by H. R. Giger’s biomechanical forms and the desolate worlds painted by Zdzisław Beksiński.
  • 8 to 12 hours of gameplay, depending on your pace.
  • A dark, atmospheric soundtrack with occasional heavier passages influenced by metal.
  • A complete, standalone story. No knowledge of the original game is required.

About the team

We spent almost eight years making Tormentum II. The core development team consists of two people: Piotr Ruszkowski and Łukasz Rutkowski, both of whom helped create the original Tormentum: Dark Sorrow.

The post Tormentum II appeared first on FitGirl Repacks.

✇FitGirl Repacks

Reap and Rush – v1.0072205 + Supporter Bundle DLC

作者 FitGirl

#7022 Reap and Rush v1.0072205 + Supporter Bundle DLC


Genres/Tags: Arcade, Shoot 'em up, Bullet hell, Top-down, 2D
Companies: Beijing Pacifier Technology, Mecrew Games
Languages: RUS/ENG/MULTI6
Original Size: 3.7 GB
Repack Size: 1.5 GB

Download Mirrors (Direct Links)

Download Mirrors (Torrent)

Discussion and (possible) future updates on CS.RIN.RU thread

Screenshots (Click to enlarge)



Repack Features

  • Based on Reap.and.Rush-TENOKE ISO release: tenoke-reap.and.rush.iso (3,929,763,840 bytes)
  • Game version: v1.0072205; Supporter Bundle DLC is included
  • 100% Lossless & MD5 Perfect: all files are identical to originals after installation
  • NOTHING ripped, NOTHING re-encoded
  • Significantly smaller archive size (compressed from 3.7 GB to 1.5 GB)
  • Installation takes 1-3 minutes (depending on your system)
  • After-install integrity check so you could make sure that everything installed properly
  • HDD space after installation: 3.6 GB
  • Language can be changed in game settings
  • Repack uses XTool library by Razor12911
  • At least 2 GB of free RAM (inc. virtual) required for installing this repack
Game Description
‘Reap and Rush’: Longer mortal lifespans have thrown the underworld into chaos, forcing its envoys down to the mortal realm to restore order. But reincarnation is broken and your boss Zhong Kui is missing. Dive into the chaos and join a hilarious, action-packed brawl across the realms!

Face endless waves of Yaoguai on a treacherous battlefield. Carve a whirlwind of destruction through the crowd and savor the simple, visceral joy of underworld envoys doing their jobs. Choose your champion from unique underworld envoy generals and dominate the battlefield.

Unleash unpredictable power by freely combining fate stones and yin spells. Use underworld currency to stock up on powerful paper crafts, forging countless synergies and combat styles. Master your unique build and discover that no two runs are ever the same.

Your duties go beyond battle. Solve the underworld’s daily chaos such as feed a hungry Yaoguai, help a misplaced god or untangle a fox spirit in his cultivation… A top-tier envoy masters both the pen and the blade. Ready to negotiate peace or clear the scene!

Game designers are committed to providing players with fresh, timely and interesting gameplay. We highly value feedback from all players and are committed to improving the game for everyone. Your valuable ideas can help us to optimize the quality of Reap and Rush.

The post Reap and Rush – v1.0072205 + Supporter Bundle DLC appeared first on FitGirl Repacks.

✇FitGirl Repacks

Boba Cafe Simulator – v1.034

作者 FitGirl

#7021 Boba Cafe Simulator v1.034


Genres/Tags: Lifestyle, First-person, 3D
Companies: Erabit, Caikemi
Languages: RUS/ENG/MULTI11
Original Size: 2.7 GB
Repack Size: 1.3 GB

Download Mirrors (Direct Links)

Download Mirrors (Torrent)

Discussion and (possible) future updates on CS.RIN.RU thread

Screenshots (Click to enlarge)



Repack Features

  • Based on Boba.Cafe.Simulator-TENOKE ISO release: tenoke-boba.cafe.simulator.iso (2,862,151,680 bytes)
  • Game version: v1.034
  • 100% Lossless & MD5 Perfect: all files are identical to originals after installation
  • NOTHING ripped, NOTHING re-encoded
  • Significantly smaller archive size (compressed from 2.7 to 1.3 GB)
  • Installation takes one minute
  • After-install integrity check so you could make sure that everything installed properly
  • HDD space after installation: 2.4 GB
  • Language can be changed in game settings
  • Repack uses XTool library by Razor12911
  • At least 2 GB of free RAM (inc. virtual) required for installing this repack
Game Description
Run a milk tea shop with up to 8 players in co-op. Craft drinks by hand, design menus and prices, place equipment, decorate your shop, handle customer flow and delivery platforms—and build your ultimate milk tea empire.

Have you ever dreamed of owning your milk tea shop?

In Boba Cafe Simulator – Reborn as the Manager of Bingbao, you and your friends become business partners, embarking on a cozy yet strategic management adventure.

Start a brand-new shop with friends—or take full control and run everything on your own in single-player mode.

The post Boba Cafe Simulator – v1.034 appeared first on FitGirl Repacks.

✇9to5Mac

T-Mobile ‘actively working’ to fix ongoing outage, iPhone SOS mode

作者 Chance Miller

Update #2, 4:15 p.m. PT: In a statement to 9to5Mac, a T-Mobile spokesperson said:

“Our teams are actively working the reports of technical challenges impacting some customers. Any service impacts are in the process of being resolved and are our highest priority. We know how critical it is to stay connected, and every available resource is focused on resolving this as quickly as possible.”

Update #1, 3 p.m. PT: Some T-Mobile users report that service is returning to their iPhones. Others, however, say the outage is ongoing.


T-Mobile is down across the United States, and many iPhone users are seeing “SOS” in their device’s status bar. Here’s everything you need to know about today’s T-Mobile outage.

✇9to5Mac

MagSafe Monday: ESR FlickLock solved my biggest complaints with the AirPods Pro 3

作者 Bradley C

I drop my AirPods Pro all the time. In fact, I dropped them at the lake yesterday. The worst part is always watching the earbuds bounce across the pavement and hoping they do not fall down a storm drain or something. I love having a case that actually stays shut when it hits the ground. The new ESR FlickLock Magnetic Case is designed specifically for the AirPods Pro 3 to solve that exact problem. It adds a secure locking lid and an incredibly strong MagSafe connection.

✇T客邦-Techbang

台積電傳2027年全面漲價10%!美亞利桑那州投資再加碼

作者 janus
台積電因應成本壓力與AI需求,傳出將於2027年啟動漲價。儘管海外擴廠支出龐大,台積電仍憑藉領先的晶圓代工技術展現強大定價權。未來晶圓代工產業將持續受惠於AI浪潮,使這次預告漲價備受市...00c61f4842f16654d1d7e5282121880b

 

根據《日經亞洲》的最新報導指出,多位產業消息人士透露,台積電(TSMC)正計畫在 2027 年將其先進與成熟製程晶圓代工服務的價格最高調漲 10%。

此次價格調整的主要驅動力,來自於多方面的營運壓力,包括原物料價格上漲、製造設備採購成本增加,以及近年來大舉在海外新建晶圓廠所衍生的高昂營運費用。此外,全球供應鏈持續面臨的重組壓力,以及針對人工智慧(AI)晶片等高算力需求所帶來的高額研發與資本支出,也都是促成此次調漲決策的重要關鍵因素。

隨著半導體市場的快速變化與前瞻技術的推進,台積電已經開始就 2027 年的產能預訂情況,與主要客戶展開新一輪的定價方案協商。雖然距離 2027 年還有一段時間,但為了確保長期投資的回報率與穩定的產能供應,提前鎖定合約與調整價格成為必要的商業策略。此舉不僅反映了晶圓代工產業在先進製程上面臨的極高成本門檻,也顯示出台積電在面對龐大產能需求時,擁有足夠的定價話語權來維持其長期的獲利結構。

亞利桑那州投資規模擴大 競爭對手步步緊逼

在海外佈局方面,根據《彭博社》稍早的報導,台積電財務長黃仁昭在接受專訪時明確表示,公司對於美國亞利桑那州廠區的建設進展感到「非常滿意」,因此決定將該地區的總投資額度大幅提升至高達 2,650 億美元。

目前,台積電在亞利桑那州的首座晶圓廠已經正式投入營運,第二座晶圓廠也即將開始進行設備搬遷作業;同時,第三座晶圓廠正在如火如荼地建設當中,而第四座晶圓廠以及該廠區首個先進封裝設施的前期準備工作也已全面啟動,展現出其深耕在地化生產的決心與超高執行力。

儘管台積電長期以來一直在先進晶片製造領域保持絕對的市場領導地位,但其競爭對手正積極尋求縮小技術與市佔率的差距。

其中,三星電子(Samsung Electronics)受惠於近期記憶體晶片市場的強勢復甦,擁有更充沛的資金投入代工技術研發;而英特爾(Intel)也未曾放棄重返榮耀的野心。面對同業的競爭壓力,財務長黃仁昭展現出強烈自信,他強調公司對現有的商業模式依然充滿信心,並直言不諱地表示:「我們不打算留下任何可能性,我們的競爭對手很強,但我們更強。」

 

總結來說,台積電預告 2027 年的漲價策略,表面上是反映通膨與建廠成本的被動反應,實則展現了其在 AI 時代下不可替代的戰略地位。

當高算力晶片成為各國與各大科技巨頭爭奪的核心資源時,擁有最高良率與最先進封裝技術的台積電,自然有底氣要求供應鏈共同分擔研發與擴產成本。未來,隨著亞利桑那州等多處海外廠區的陸續完工與投產,全球半導體版圖將迎來實質性的重塑,而台積電能否在競爭對手的夾擊下繼續保持「更強」的姿態,將是業界與投資人持續關注的焦點。

 

 

加入T客邦Facebook粉絲團
❌