OpenAI因担忧推迟发布AI检测工具

币界网Published on 2024-08-05Last updated on 2024-08-05

币界网报道:

OpenAI推迟了其备受期待的工具的发布,该工具旨在检测其ChatGPT模型生成的文本。据报道,由于可能影响用户体验和生态系统的问题,发布已被推迟。

《华尔街日报》称,OpenAI在过去的一年里一直在准备发布该工具,但由于风险和各种问题,它还没有这样做。推迟工具发布的选择似乎也与用户的输入有关。在OpenAI进行的一项民意调查中,他们发现,近三分之一的忠实ChatGPT用户可能会被反作弊技术的引入所吓倒。

反作弊工具的功能

根据民主与技术中心的数据,59%的中学和高中教师认为学生在作业中使用了人工智能,比上一年有所增加。

反作弊工具的工作原理是在人工智能编写的文本中嵌入水印。即使这个水印肉眼不可见,OpenAI的检测技术也能检测到它。该工具提供了一个分数,表示给定文本由ChatGPT编写的概率。据说这种水印技术几乎是完美的,因为内部文件显示它有99。当ChatGPT产生大量内容时,效率为9%。

然而,关于该工具提出了一些问题。一个主要问题是,这可能对那些不是英语母语的人不利。有些人担心水印可能没有达到应有的效率,或者可能会将英语技能较差的作者的其他作品标记为抄袭。此外,批评者表示担心,通过翻译或插入和删除表情符号,水印可能会被擦除或操纵。

另一个问题是应该允许谁使用检测工具。限制访问可能会降低水印方法的适用性;然而,允许自由访问可能会使该方法无效。

其他科技巨头也开发了类似的解决方案

尽管OpenAI尚未实现文本反作弊工具,但该公司已经开始在音频和视频输出中应用水印。这些多媒体格式被该公司视为优先事项,因为人工智能生成的deepfakes比基于文本的内容具有更高的风险。

其他科技公司也在开发类似的解决方案。例如,谷歌还推出了自己的水印解决方案SynthID,目前仍处于测试阶段。然而,OpenAI将音频和视觉水印置于文本之上,因为前者的影响更为严重,可能发生在选举年等关键时期。

这一消息发布之际,埃隆·马斯克(Elon Musk)再次出庭起诉OpenAI,指控该公司及其创始人萨姆·奥特曼(Sam Altman)和格雷格·布罗克曼(Greg Brockman)偏离了创造人工智能以改善人类的最初目标。

Trending Cryptos

Related Reads

Borrowed Faith: How Much of the Bitcoin ETF Flows Are Real Money

"Rented Faith: How Much of Bitcoin ETF Flows Are Real Money?" Bitcoin ETF inflows are often seen as a barometer of institutional conviction. However, week-to-week analysis reveals they are primarily driven by a hidden arbitrage trade rather than directional bullishness. This is the cash-and-carry trade: buying the ETF while simultaneously shorting Bitcoin futures on the CME to lock in the price difference (basis). This delta-neutral activity registers as ETF inflows but reflects a rate-seeking, not price-betting, strategy. Data shows weekly ETF flow volatility is closely tied to hedge fund ("leveraged funds") short positions on CME futures, with a correlation of 0.70. About half of weekly flow variation can be explained by this single factor. In contrast, Bitcoin's weekly price changes have no statistically significant power to predict flows. Crucially, while this arbitrage trade dominates weekly *fluctuations*, it is not the main component of the cumulative *stock*. Of the ~$55 billion total net inflow, the estimated net arbitrage position is only about $1 billion. The vast majority is steady, directional buying averaging ~$400 million per week. Thus, ETF flows overstate the *volatility* of belief, not its *level*. The "rented" arbitrage capital churns, while "owned" directional capital forms the bedrock. This arbitrage trade has been unwinding for two years, with hedge fund shorts peaking at ~$14 billion in late 2024 and falling to ~$4.5 billion. Recent outflows align with basis compression, signaling the trade's exit, not a loss of faith. For Ethereum ETFs, the same dynamic is weaker due to negative carry from forgone staking yield. The key takeaway: To interpret ETF flows, watch the basis vs. Treasury yields and CME hedge fund net shorts. They reveal how much of the weekly "demand" headline is driven by rented, rate-seeking capital versus real conviction.

链捕手2m ago

Borrowed Faith: How Much of the Bitcoin ETF Flows Are Real Money

链捕手2m ago

Two Giants' Credit Expansion: Loan Balances of $9.9 Billion vs. $14.6 Billion, Brazil Emerges as the Main Battlefield

Title: Two Giants "Credit" Surge: Loan Balances of 99 Billion vs. 146 Billion USD, Brazil Emerges as Main Battlefield Summary: The article compares the rapid expansion of credit businesses by two major e-commerce and fintech players, Sea (via Monee) and Mercado Libre (via Mercado Pago), in overseas markets like Southeast Asia and Latin America, contrasting with a slowing domestic Chinese credit market. Using Q1 2026 financial data, it highlights their significant growth. Sea's Monee reached a loan balance of $99 billion, up 71% year-over-year (YoY), contributing 17.5% to Sea's total revenue. Mercado Pago's loan balance hit $146 billion, up 87% YoY, contributing 45% to its parent company's revenue. Both maintained stable risk metrics (e.g., Monee's 90+ day NPL at 1.1%) despite rapid scaling. Brazil is identified as a key and accelerating growth market for both. Sea's Brazilian operations saw loan volumes exceed $10 billion, growing 250% YoY, with SPayLater GMV penetration still low (~10%) indicating high potential. Sea also secured a key Brazilian financial credit license (SCFI). Mercado Libre's Brazil segment contributed over half (54%) of total group revenue, with its credit business there generating $11.24 billion in revenue, up 89% YoY and accounting for 12.7% of global revenue. Mercado Pago's credit portfolio, especially credit cards (46% of loans, +105% YoY), is a strategic focus, described as crucial as building logistics was a decade ago. Its net interest margin after loss (NIMAL) remains high at 17.8%. The article concludes that while Brazil presents immense opportunities, the success is largely driven by these integrated "e-commerce + fintech" giants with proprietary transaction data and ecosystems, making it challenging for standalone fintech players to compete effectively.

链捕手23m ago

Two Giants' Credit Expansion: Loan Balances of $9.9 Billion vs. $14.6 Billion, Brazil Emerges as the Main Battlefield

链捕手23m ago

Research Report Analysis: Is Intel Making a Comeback with Apple? Bernstein's Calculations Show the Right Direction, but the Price Is Already Overvalued

Bernstein analyst Stacy A. Rasgon published a report on June 18 regarding Intel, assessing the potential impact of recent political support for a US-based PC chip design and manufacturing collaboration between Apple and Intel. The report views this as a significant signal for the foundry landscape shift but concludes the initial financial contribution would be minimal. Key conclusions: 1) An Apple deal is seen as a small-scale "proof of concept." Even if Intel wins 40% of Apple's premium notebook chip orders (~5 million units/year), Bernstein estimates it would generate only about $500M in annual revenue and ~$0.03 EPS, negligible against Intel's ~$55B revenue. 2) Political encouragement is not equivalent to enforceable mandates. Winning orders ultimately depends on Intel demonstrating competitive technology (like its 18A node), cost, and reliable supply. 3) The path from validation to large-scale production involves significant challenges, capital investment, and time. Due to these uncertainties, Bernstein maintains a Market-Perform (Hold) rating with a $100 price target, implying potential downside from the ~$121.10 price at the report date. The analysis highlights the tension between near-term validation value—serving as a crucial trust signal for Intel's foundry ambitions and US supply chain resilience—and the long-term opportunity to attract larger cloud and AI chip customers. The investment thesis hinges on successful 18A execution and sustained policy support, not on immediate financial gains from Apple.

marsbit48m ago

Research Report Analysis: Is Intel Making a Comeback with Apple? Bernstein's Calculations Show the Right Direction, but the Price Is Already Overvalued

marsbit48m ago

Trading

Spot
Futures

Hot Articles

Discussions

Welcome to the HTX Community. Here, you can stay informed about the latest platform developments and gain access to professional market insights. Users' opinions on the price of AI (AI) are presented below.

活动图片