桃園市府於兒童節連假期間邀請親子家庭出門走走，開箱七座全新「P@RK園宇宙」特色公園、14家觀光工廠、5家食品品牌

4月到觀光工廠參加限定活動，帶「桃園好『ㄅㄠ∨』包」回家(桃園購物商業消費系列活動)。

https://fstry.pse.is/7cpqxf

—— 以上為 Firstory Podcast 廣告 ——

-----------------------------------
🎧 通勤學英語 VIP 專區 & 🚀 線上課程 
-----------------------------------
💡 想要更自然的提升英語力？加入VIP會員，獲得專屬內容與優惠！

• 立即加入VIP方案 →https://open.firstory.me/join/15minstoday
• VIP訂閱常見問題 →https://15minsengcafe.pse.is/5cjptb

🔥 社會人核心英語有聲書課程→https://15minsengcafe.pse.is/554esm

-------------------------------
🌎 15Mins.Today 相關連結
-------------------------------
💬 你的想法很重要！留言分享 → 留言連結

Web：www.15mins.today
YouTube：https://15minsengcafe.pse.is/3rhuuy
商業合作/贊助來信：[email protected]
意見回覆 : [email protected]

--------------------------------------------
📜 單集逐字稿 & 內容摘要（播放器字數有限，完整逐字稿請到官網查看）
--------------------------------------------

國際時事跟讀 Ep.K963: AI Giants Clash: Musk's Grok-3 Disrupts the Field

Highlights 主題摘要：

• Elon Musk's new AI model Grok-3 is very good at math and science problems, competing with other top AI systems from big technology companies.
• There is a personal fight between Musk and OpenAI's leader Sam Altman that started when Musk left the company he helped create.
• The two AI systems show different ideas about artificial intelligence - one focuses on real-time data while the other is better at language and creative tasks.

Elon Musk's artificial intelligence startup xAI has launched Grok-3, establishing a powerful new competitor in the AI landscape. Released in February 2025, this latest model marks a dramatic improvement over its predecessor, having been trained with ten times more computing power. According to benchmark tests, Grok-3 excels in mathematical reasoning, science, and coding—outperforming models from Google, Anthropic, and Meta on the Artificial Analysis Quality Index. While it still ranks below OpenAI's o3 model and DeepSeek-R1 in overall capabilities, its specialized strengths have captured industry attention.

伊隆．馬斯克的人工智慧新創公司 xAI 推出了 Grok-3，在 AI 領域建立了一個強大的新競爭者。這款於 2025 年 2 月發布的最新模型較其前身有顯著提升，採用了十倍於前代的運算能力進行訓練。根據基準測試，Grok-3 在數學推理、科學和程式編碼方面表現出色——在人工分析品質指數（Artificial Analysis Quality Index）上超越了來自 Google, Anthropic 和 Meta 的模型。儘管在整體能力上仍排在 OpenAI 的 o3 模型和 DeepSeek-R1 之後，但其專業優勢已吸引了業界的關注。

The model's release intensifies the ongoing rivalry between Musk and OpenAI CEO Sam Altman. Their conflict has deep roots—Musk co-founded OpenAI in 2015 but departed its board in 2018, citing conflicts with his AI work at Tesla. Since then, he has become one of OpenAI's fiercest critics. In 2024, Musk sued the organization, alleging it had abandoned its original nonprofit mission in favor of commercial interests, particularly through its Microsoft partnership. The tension escalated dramatically in early 2025 when Musk attempted to acquire OpenAI's nonprofit arm for $97.4 billion. Altman rejected this offer and countered with a provocative $9.74 billion bid to purchase X, Musk's social media platform—just a fraction of Musk's offer and clearly intended as a pointed rebuke.

這款模型的發布加劇了馬斯克與 OpenAI 執行長山姆．阿特曼之間的持續競爭。他們的衝突由來已久——馬斯克於 2015 年共同創立了 OpenAI，但在 2018 年離開了董事會，理由是與他在特斯拉的 AI 工作存在衝突。從那時起，他成為了 OpenAI 最激烈的批評者之一。2024 年，馬斯克起訴了該組織，指控其放棄了原本的非營利使命，轉而追求商業利益，特別是通過與微軟的合作這點。2025 年初，緊張局勢急劇升溫，當時馬斯克試圖以 974 億美元收購 OpenAI 的非營利部門。阿特曼拒絕了這個提議，並以 97.4 億美元反要購買馬斯克的社交媒體平台 X——這僅是馬斯克出價的一小部分，明顯意在給以尖銳的反擊。

The technical differences between Grok-3 and GPT-4 reflect divergent visions for AI development. Grok-3 emphasizes real-time data access and social media integration, making it particularly effective for technical problem-solving. Its innovative features include DeepSearch, which pulls live information from the web and X platform, and Big Brain Mode, designed for complex reasoning tasks. In direct comparisons, Grok-3 outscored GPT-4o in the AIME 2025 mathematics competition and on the GPQA science benchmark, cementing its reputation for STEM excellence.

Grok-3 和 GPT-4 之間的技術差異反映了 AI 發展的不同願景。Grok-3 強調即時數據訪問和社交媒體整合，使其在技術問題解決方面特別有效。其創新功能包括 DeepSearch，可從網絡和 X 平台獲取即時信息，以及為複雜推理任務設計的進階推理模式（Big Brain Mode）。在直接比較中，Grok-3 在 2025 年美國數學邀請賽（AIME 2025）和 GPQA 科學基準測試中的得分超過了 GPT-4o，鞏固了其在 STEM （理科）領域的卓越聲譽。

GPT-4, meanwhile, continues to dominate in language understanding, creative content generation, and multimodal processing. It delivers more natural and structured responses across various topics and maintains widespread adoption in business, education, and content creation. OpenAI emphasizes its commitment to responsible AI guidelines and accuracy, positioning GPT-4 as the more measured, versatile option for general applications.

與此同時，GPT-4 在語言理解、創意內容生成和多模態處理方面繼續占據主導地位。它能夠在各種主題上提供更自然、更有結構的回應，並在商業、教育和內容創作領域保持廣泛應用。OpenAI 強調其負責任的 AI 準則和對精確性的承諾，將 GPT-4 定位為一個更為穩健、多用途的通用應用選擇。

The contrasting approaches of these AI systems reflect a fundamental question about artificial intelligence's future: should it function primarily as a corporate asset or an open tool? As xAI's Grok-3 continues to evolve—with Musk promising improvements "literally every day"—and OpenAI refines its models, this philosophical divide may shape the entire AI industry. For users, the competition delivers more options and specialized capabilities, even as the rivalry between tech billionaires turns increasingly personal and public.

這些 AI 系統的對比方法反映了關於人工智慧未來的一個根本問題：它應該主要作為企業資產還是開放工具運作？隨著 xAI 的 Grok-3 不斷發展——馬斯克承諾「幾乎每天」都有改進——而 OpenAI 也在完善其模型，這種哲學上的分歧可能會塑造整個 AI 產業。對用戶來說，這場競爭不僅提供了更多選擇和專業功能，同時科技億萬富豪們之間的競爭也日益個人化且公開化。

Keyword Drills 關鍵字：

1. Predecessor (Pred-e-ces-sor) [noun. someone or something that came before]: This latest model marks a dramatic improvement over its predecessor, having been trained with ten times more computing power.
2. Nonprofit (Non-prof-it) [noun. an organization that uses its money to help people, not to make money]: In 2024, Musk sued the organization, alleging it had abandoned its original nonprofit mission in favor of commercial interests.
3. Mathematics (Math-e-mat-ics) [noun. the study of numbers, shapes and patterns]: In direct comparisons, Grok-3 outscored GPT-4o in the AIME 2025 mathematics competition and on the GPQA science benchmark.
4. Versatile (Ver-sa-tile) [adjective. able to do many different things well]: OpenAI emphasizes its commitment to responsible AI guidelines and accuracy, positioning GPT-4 as the more measured, versatile option for general applications.
5. Rivalry (Ri-val-ry) [noun. competition between people or groups who want the same thing]: The model's release intensifies the ongoing rivalry between Musk and OpenAI CEO Sam Altman.

Reference article:
1. https://www.nbcnews.com/tech/tech-news/musk-xai-releases-grok-3-rcna192670
2. https://www.ccn.com/education/crypto/grok-3-vs-gpt-4-key-differences-explained/