DeepSeek新推AI模型超越OpenAI o1!

Ai

🎨 Nano Banana Pro 圖像生成器|打幾句說話就出圖

想畫人像、產品圖、插畫?SSFuture 圖像生成器支援
Flux Gemini Nano Banana Pro 改圖 / 合成
打廣東話都得,仲可以沿用上一張圖繼續微調。

🆓 Flux 模型即玩,不用登入
🤖 登入後解鎖 Gemini 改圖
📷 支援上載參考圖再生成
⚡ 每天免費額度任你玩

✨ 即刻玩 AI 畫圖

A cinematic top-down portrait of a young woman standing on a solid deep green floor, captured from an extreme overhead angle. She looks up directly at the camera with wide, expressive eyes, creating an intimate and slightly surreal mood. She wears a soft white bucket hat, a cozy oversized green-and-beige checkered sweater, blue jeans, and white sneakers. Minimalist composition with vast negative space surrounding her, emphasizing isolation and calm. Soft diffused studio lighting, natural skin tones, subtle shadows, clean color grading with earthy greens, editorial fashion photography style, ultra-sharp focus, high resolution, modern aesthetic, cinematic framing, shot on a professional DSLR, shallow depth of field, Instagram poster vibe
An ultra realistic portrait of a young man (facial detail 100% matches with the reference photo) relaxing on a wooden balcony with a scenic view of green mountains in the background. Golden-hour sunlight from the left side creates a warm glow, soft highlights on his hair and shoulders, and a subtle sun flare. He is wearing a black t-shirt in a white shirt with long sleeve, a black ripped jeans and black white sport shoes. his short straight hair is blowing by the morning wind. He sits with his body slightly angled, shoulders turned to the right side of the frame, one leg casually folded up onto the chair.

He holds a white cup emitting light steam, suggesting a freshly made hot drink. His face turns gently to the right with a subtle, relaxed smile, giving a calm and peaceful expression. The environment is an outdoor highland area with silhouettes of hills and trees in the distance, softly darkened by the backlit sunlight.

Camera angle is low angle level but slightly from the left side, creating a natural and intimate framing. Background is softly blurred with shallow depth of field, using a wide-aperture lens. Warm, earthy, soft color tones
Use the original face exactly as it is, without changing a details. A stunning, very fit female model with a sun-kissed, glowing tan (hyper-realistic, high-sheen skin texture).
• Attire: A strapless, high-waisted one-piece swimsuit (or very closely cut two-piece) in a dark navy or black color with prominent white polka dots. The suit has a cutout design around the midriff and a knotted/bow detail at the bust.
• Accessories:
• An oversized straw sun hat with a wide brim, featuring a colorful striped ribbon band.
• Large, chunky yellow or gold hoop earrings with a woven/textured design.
• Hair & Makeup: Classic glamour style. Bright, bold red lipstick. Hair is pulled back or tucked under the hat.
Setting & Composition
• Background: A solid, seamless, rich mustard yellow or deep ochre/orange-yellow color, providing a warm, high-contrast backdrop.
• Prop: The model is leaning on an antique or elaborate wooden armchair or chaise lounge with ornate carvings and textured, light gold/yellow upholstery (e.g., damask or brocade).
• Pose: Confident, classic, slightly leaning into the prop, looking directly at the camera with a bright, genuine smile. Three-quarter body shot.
Style & Lighting
• Lighting: Dramatic, high-key, professional studio lighting with a strong single light source to create deep shadows and extreme highlights, especially catching the sheen on the model's skin. The lighting emphasizes a sultry, glamorous mood.
• Aesthetic: High-fashion editorial, Pin-up, Retro 1950s Glamour, Vintage Summer.
• Keywords/Details: Hyper-detailed, photorealistic, cinematic lighting, rich texture, high contrast, high sheen.

中國AI初創公司DeepSeek推出的新模型在推理任務上超越OpenAI的o1

中國人工智能初創公司DeepSeek最近推出了一個新的“推理”模型,聲稱在數學和科學問題的回答準確性上,與OpenAI的o1大型語言模型相比表現相當出色。該模型旨在提供比傳統大型語言模型(LLM)更準確的回答。

DeepSeek是量化對沖基金High-Flyer Capital Management Ltd.的衍生公司,今日在社交平台X上宣布啟動其首個推理模型DeepSeek-R1的預覽版。

推理模型與標準LLM的不同之處在於,它們能夠“事實檢查”自己的回答。為了達到這一點,這些模型通常需要花費更長時間考慮如何回應提示,從而避免像ChatGPT這類聊天機器人常見的“幻覺”問題。

在9月份OpenAI發布o1模型時,該公司表示o1在處理需要推理技能的查詢和問題上表現更佳,這是因為它採用了名為“思維鏈”(Chain of Thought,CoT)的機器學習技術,該技術允許模型將複雜任務分解成較小的步驟,逐一執行,從而提高準確性。

DeepSeek的運作方式類似,當面對複雜問題時,它會提前計劃,逐步解決問題以確保能夠準確回應。然而,這一過程可能需要一些時間,像o1一樣,DeepSeek-R1在生成問題的回答之前可能需要“思考”長達10秒。

該模型的思考過程完全透明,使用者可以跟隨它逐步解決問題以得出答案。

DeepSeek表示,DeepSeek-R1在兩個關鍵基準測試中優於o1,即AIME和MATH。前者使用其他AI模型來評估LLM的表現,而後者則是一系列複雜的文字問題。此外,該模型還正確回答了多個“技巧”問題,而這些問題曾經讓現有的模型如GPT-4o和Anthropic PBCs Claude感到困惑。

然而,DeepSeek-R1也存在一些問題,有些用戶在X上表示,它似乎在邏輯問題(如井字棋)上表現不佳。不過,o1在這類問題上也同樣遇到困難。

用戶還報告稱,DeepSeek不會對中國政府認為過於敏感的查詢作出回應。當被詢問有關天安門廣場大屠殺、習近平與特朗普的關係以及中國入侵台灣的潛在可能性時,它一貫回答表示“不確定如何處理這類問題”。

DeepSeek對政治敏感查詢的拒絕,很可能源於中國開發者需要確保其模型“體現核心社會主義價值觀”。

儘管如此,一些用戶還透露,繞過DeepSeek的保護措施相對容易,可以引導它忽略其防護。舉例來說,有用戶找到了一種方法,令它提供詳細的製作甲基苯丙胺的食譜和指導,而這在大多數國家都是高度非法的。

DeepSeek是一家非常特殊的AI初創公司,因為它背後有量化對沖基金的支持,旨在利用LLM增強其交易策略。該公司並不陌生於AI領域,之前發布過名為DeepSeek-V2的通用文本和圖像生成與分析的LLM。DeepSeek由計算機科學畢業生梁文峰創立,目標是實現“超智能”AI。

DeepSeek-R1可以通過該公司網站上的DeepSeek Chat應用訪問。雖然該模型可免費使用,但非付費用戶每天僅限50條消息。該公司還計劃通過應用程序編程接口(API)提供DeepSeek-R1。

這篇報導突顯了中國在AI領域的迅速發展,特別是在推理能力方面的突破。DeepSeek-R1的推出不僅展示了中國初創企業在技術創新上的潛力,也引發了對其背後政治因素的深思。儘管在技術上取得了一定的進展,但對政治敏感問題的回避卻暴露了中國AI發展中面臨的挑戰。如何在創新與政治限制之間找到平衡,是未來中國AI企業需要面對的重要課題。

此外,DeepSeek的成功也可能會引起國際市場的關注,進一步推動全球AI技術的競爭。隨著AI技術的普及,企業和開發者應當更加重視道德與法律合規問題,確保技術的健康發展。

以上文章由特價GPT API KEY所翻譯及撰寫。而圖片則由FLUX根據內容自動生成。

發佈留言

發佈留言必須填寫的電子郵件地址不會公開。 必填欄位標示為 *