OpenAI的未来展望:Strawberry、Orion和GPT下一代计划

marsbitPublished on 2024-09-03Last updated on 2024-09-04

OpenAI即将发布两款可能重新定义机器学习领域的革命性模型,代号分别为Strawberry和Orion。这些项目的目标是将AI的能力推向新的高度,特别是在推理、解决问题和语言处理方面,朝着人工通用智能(AGI)更进一步。

Strawberry(之前被称为 Q* 或 Q-Star)似乎不仅仅是一个聊天机器人;它的重点是展示 AI 推理能力的显著飞跃。知情人士向路透社和 The Information 等媒体透露,该项目在解决复杂数学问题和增强逻辑分析方面表现出色。

另一方面,Orion 被定位为 OpenAI 的下一代旗舰语言模型,可能是 GPT-4 的继任者。它旨在在语言理解和生成方面超越其前身,并具备处理多模态输入(包括文本、图像和视频)的能力。

这两个项目引起了美国国家安全官员的关注,突显了它们可能具有的战略重要性。在 OpenAI 尽管收入增长显著的情况下,仍在继续筹集资金,这可能与开发和训练这些先进模型所需的高昂成本有关。

Strawberry与推理能力

尽管网上不断有关于“Strawberry项目”的猜测,OpenAI并未正式发布任何相关信息。然而,据称泄露的信息表明,Strawberry在复杂推理方面表现出色。

据称,与传统模型提供快速响应不同,Strawberry采用了研究人员称为“系统2思维”的方法,能够花时间仔细推理并解决问题,而不仅仅是通过预测更长的词语序列来完成响应。据路透社报道,该模型在MATH基准测试中取得了90%以上的成绩,表现令人印象深刻。

另一个预计由 Strawberry 带来的关键创新是其生成高质量合成训练数据的能力。这一功能解决了 AI 开发中的一个重要挑战:缺乏多样且高质量的训练数据。如果这一点属实,Strawberry 不仅能提升自身能力,还为更先进的模型如 Orion 铺平了道路。

考虑到 OpenAI 已经获取的大量数据,以及用户日益重视隐私而不愿向 AI 提供数据的趋势,这一功能可能在未来 AI 模型的质量中发挥重要作用,就像如今一些用户利用 Stable Diffusion 生成的图像训练自己的定制模型一样。

然而,Strawberry 的这种深思熟虑的处理方式可能在实时应用中面临挑战。据报道,OpenAI 的研究人员正在研究如何“蒸馏” Strawberry 的能力,基本上是降低其质量,以便用户能够在低计算成本下进行大量推理。

即便如此,将 Strawberry 的技术整合到面向消费者的产品(如 ChatGPT)中,可能会显著提升 OpenAI 训练新模型的方式。然而,更有可能的是,OpenAI 将利用 Strawberry 作为训练新模型的基础,而不是直接向消费者广泛提供。

Project Orion 或 GPT 下一代

Project Orion 作为 OpenAI 对 GPT-4 的雄心勃勃的继任者,旨在设定语言 AI 的新标准。根据 OpenAI 日本公司 CEO 永崎忠雄的最新演讲,项目可能被命名为 GPT 下一代(GPT Next)。利用来自 Project Strawberry 的先进技术,Orion 设计不仅在自然语言处理方面表现卓越,还扩展到多模态能力。

OpenAI声称,下一代AI模型的进步将不会是渐进的。

根据IT Media的报道,OpenAI日本公司CEO永崎忠雄在2024年日本KDDI峰会上表示:“即将推出的AI模型可能被称为‘GPT Next’,其发展速度将是前几代模型的近100倍。与传统软件不同,AI技术呈指数级增长。因此,我们希望尽快支持一个将AI全面整合的世界的诞生。”

将 Strawberry 生成的数据用于训练 Orion 对 OpenAI 来说是一项技术优势。然而,这种技术需要谨慎使用。研究人员已经证明,当模型在过多的合成数据上进行训练时,其性能会开始下降。因此,找到一个合适的平衡点,使 Strawberry 可以增强 Orion 的能力而不影响其准确性,对于 OpenAI 保持竞争力至关重要。

Orion 的原生多模态功能也将是一个显著的进步。据《The Information》报道,该模型正在开发中,旨在无缝整合文本、图像,甚至视频输入和输出,为 ChatGPT 用户带来全新的可能性,并使公司与谷歌的 Gemini 直接竞争。后者可以处理长达 2 小时的视频输入。

这是用户在使用 ChatGPT 或 OpenAI 的 API Playground 时将与之交互的模型。

Orion 的开发与 OpenAI 的整体战略相一致,旨在保持其在日益拥挤的 AI 领域中的竞争优势。随着开源模型(如 Meta 的 LLaMA-3.1)和先进模型(如 Claude 或 Gemini)的快速进展,Orion 基本上是 OpenAI 保持领先地位的举措。

Trending Cryptos

Related Reads

BIS Report Compliance Observations: The True Risks of Stablecoins Go Beyond 'De-pegging'

The BIS report, "Anchoring trust in money: innovation beyond stablecoins," highlights that the primary risks of stablecoins extend beyond potential de-pegging. It argues that the core challenge is whether stablecoins can be integrated into a financial system that is identifiable, monitorable, accountable, and regulatable. While acknowledging efficiency gains like faster payments and programmability, BIS emphasizes that money requires an institutional framework—including legal certainty, liquidity support, and financial integrity controls—which many stablecoins currently lack. The report details compliance risks, noting that while blockchain transactions are transparent, address visibility does not equate to identity or purpose clarity. This creates a systemic risk as pseudonymity, non-custodial wallets, and cross-chain bridges can undermine AML/CFT controls. Furthermore, these risks can spill over into the traditional financial system through on- and off-ramps. The future direction, per BIS, is not to prohibit innovation but to embed regulatory rules—such as identity verification and transaction screening—directly into the technological infrastructure of tokenized finance. The key takeaway for compliance is that any new financial instrument must clearly address questions of customer identification, transaction monitoring, accountability, and cross-border rule consistency to be viable as a mainstream payment tool.

marsbit55m ago

BIS Report Compliance Observations: The True Risks of Stablecoins Go Beyond 'De-pegging'

marsbit55m ago

When US Giants Collectively "Defect" to Chinese AI Models

When Silicon Valley Giants Turn to Chinese AI Models to Cut Costs A surprising trend is emerging: major U.S. tech companies are significantly reducing AI costs by switching to Chinese models. Coinbase, the largest U.S. cryptocurrency exchange, reportedly halved its AI spending after migrating to China's GLM-5.2 and Kimi 2.7 models, despite increasing usage. They achieved this through a sophisticated three-part strategy: implementing an automatic routing system to select the most cost-effective model per task, boosting cache hit rates from 5% to 60% to reuse computations, and employing "context engineering" to provide AI with more precise, less cluttered information. They are not alone. AI startup Lindy switched from Claude to DeepSeek, saving millions, while Snowflake's tests found GLM-5.2 solved 66% of coding tasks compared to Claude Opus's 67%—but at a fraction of the cost (output pricing is 5-7 times lower). While the top Western models may offer slightly better stability, the massive price differential is leading many businesses to reconsider their value proposition. This shift signals a deeper change in the AI industry, moving beyond pure performance benchmarks to a fierce cost competition. As pressure mounts, even OpenAI and Anthropic have begun slashing prices. For users, this means more choices, lower costs, and a crucial lesson: using multiple models based on task complexity, optimizing with caching, and keeping contexts lean are now key to leveraging AI efficiently and affordably.

marsbit1h ago

When US Giants Collectively "Defect" to Chinese AI Models

marsbit1h ago

BIS Report Compliance Watch: The Real Risks of Stablecoins Are Not Just 'De-pegging'

BIS Report Compliance Observations: The real risks of stablecoins go beyond "depegging" The BIS report "Anchoring trust in money: innovation beyond stablecoins" argues that while stablecoins and tokenization offer efficiency gains, their primary risk lies in fitting into an identifiable, monitorable, accountable, and regulatable financial system. Money's trust stems not just from technology but from institutional arrangements: a common unit of account, guaranteed redemption at par, liquidity support, regulatory frameworks, and financial integrity requirements. Stablecoins, operating on permissionless blockchains with pseudo-anonymity and non-custodial wallets, create systemic compliance gaps: unclear customer identity, incomplete fund origins, unexplained transaction purposes, fragmented cross-chain paths, and ambiguous liability. On-chain transparency does not equal compliance transparency. Public addresses don't reveal identity or intent. While blockchain analytics aid law enforcement, they cannot replace routine, large-scale AML/CFT controls. Effective compliance requires a closed-loop process encompassing customer onboarding, transaction monitoring, investigation, reporting, and audit. Stablecoin risks are not confined to the blockchain; they re-enter the traditional financial system via on/off-ramps, exchanges, and payment institutions. This forces banks to monitor client accounts for activity linked to virtual assets. The future direction is not to prohibit innovation but to embed rules into the technology. Tokenized finance should integrate with the existing two-tier monetary system, embedding compliance—like customer identification, pre-transaction screening, and auditable data trails—directly into the transaction flow. For compliance professionals, the key takeaway is that any new financial instrument must answer core questions: Who identifies the customer? Who monitors transactions? Who handles exceptions? Who is liable? Compliance is not the antithesis of innovation but the essential infrastructure for its sustainable growth.

链捕手1h ago

BIS Report Compliance Watch: The Real Risks of Stablecoins Are Not Just 'De-pegging'

链捕手1h ago

When American Giants 'Defect' to Chinese AI Models

Summary: The trend of major U.S. technology firms adopting more cost-effective Chinese AI models is gaining momentum. A prime example is Coinbase, the largest U.S. cryptocurrency exchange, which reportedly halved its AI expenditure by switching to Chinese models GLM-5.2 and Kimi 2.7, while its usage volume increased. This was achieved through a sophisticated cost-saving system featuring intelligent model routing (selecting the most suitable model per task), dramatically improving cache hit rates from 5% to 60%, and implementing "Context Engineering" to streamline prompts. This shift is not isolated. Other companies like the AI startup Lindy and data cloud firm Snowflake are making similar moves, drawn by the significant price disparity. For instance, GLM-5.2 costs $1.40/$4.40 per million tokens (input/output), compared to $5/$25 for Claude Opus 4.7. While top Western models may offer slightly higher stability or speed in complex tasks, the performance gap is narrowing, making the price difference harder to justify for many enterprise use cases. The implications are significant for both businesses and individual users. It highlights the importance of a multi-model strategy based on task requirements, the value of caching and reusing outputs, and the effectiveness of providing concise context. Ultimately, this migration signals a potential reshaping of the AI industry's pricing model, moving competition from pure performance benchmarks to practical cost-effectiveness, with increased choice and downward price pressure benefiting end-users.

链捕手1h ago

When American Giants 'Defect' to Chinese AI Models

链捕手1h ago

Trading

Spot

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