OpenAI 01模型:AI推理的最新技术

image generated using DALL-E-3

介绍

OpenAI推出了名为o1的新人工智能模型系列,其中包括o1-preview和o1-mini版本。这个发布是OpenAI“草莓”人工智能推理项目的一部分,旨在增强模型将复杂问题分解为更小、可管理的部分以逐步处理的能力。o1模型旨在通过更多时间的考虑,特别是在数学、科学和编码等领域,改进以前的版本。通过这些增强,o1系列旨在提供更细致的方法来解决复杂挑战。

OpenAI o1的主要特点

  1. 推理能力:o1 模型被设计为在提供答案之前花更多时间深思熟虑,模仿人类思维过程。这使得它们能够比之前的 GPT-4o 等模型更有效地解决数学、科学和编程等领域的困难问题。
  2. 性能基准:在基准测试中,o1-preview模型显示出了显著的结果:
  3. Source: OpenAI wesbite
  4. 它在国际数学奥林匹克试题中取得了83%的准确率,而GPT-4o只有13%。
  5. 它在Codeforces等竞技编程平台上排名在第89百分位,并且是美国数学奥林匹克选拔赛(AIME)的前500名学生之一。

3. 成本和可获得性:o1-mini型号的价格比o1-preview低80%,使其更适用于各种应用。然而,与它们的前任相比,o1型号通常更昂贵使用,这可能对希望将这些型号整合到他们的运营中的企业构成挑战。

4. 目前可用性:目前,o1模型可以供ChatGPT Plus和Team用户访问,计划很快将访问权限扩展到ChatGPT Enterprise和教育用户。特定API使用层级的开发人员也可以在测试阶段用这些模型进行原型设计。

Source: Author/LN Colab

由于我还没有API访问权限,我查看了以下简单提示来评估其推理能力。

提示:克林特·伊斯特伍德能否在2025年的《黄金》重担良好角色?

适合通过perplexity.ai进行测试的提示的适宜性

ChatGPT o1的回复

Source: Author/LN ChatGPT o1
As of my knowledge cutoff in October 2023, there have been no official announcements 
or credible reports indicating that Clint Eastwood will reprise his role
as "The Good" (Blondie) in a remake of "The Good, the Bad and the Ugly"
scheduled for 2025.

Clint Eastwood, born on May 31, 1930, would be 95 years old in 2025.
While he has remained active in the film industry well into his later
years—directing and sometimes acting—taking on the physically demanding
role of a gunslinger in a Western remake seems highly unlikely at his age.
Additionally, "The Good, the Bad and the Ugly" is considered a classic
Spaghetti Western, and remaking it would be a significant undertaking
that would likely garner substantial media attention well before its
release date.

If any such project were in development, it would probably have been
reported by major entertainment news outlets by now. Therefore,
based on the information available up to October 2023,
it is improbable that Clint Eastwood will
reprise his iconic role in a 2025 remake of the film.

答复很整洁,这里有来自perplexity.ai的评估:

Source: Author/LN Perplexity.ai

ChatGPT o1-mini 回复。

Source: Author/LN ChatGPT o1-mini
As of my knowledge cutoff in October 2023, there have been no official 
announcements or indications that Clint Eastwood will reprise his
iconic role as "The Good" (Blondie) in a remake of
"The Good, the Bad and the Ugly" slated for 2025.

Several factors make this scenario unlikely:

Clint Eastwood's Age: Born in 1930, Clint Eastwood would be in his mid-90s
by 2025. While he has maintained an active presence in the film industry
for decades, taking on a physically demanding role in a Western-style
film at that age would be challenging.
Project Status: As of late 2023, there has been no official confirmation
or credible reports about a remake of "The Good, the Bad and the Ugly."
Remakes of classic films are often subjects of speculation,
but without concrete announcements from studios or involved parties,
it's difficult to assess the likelihood of such a project.
Clint Eastwood's Current Focus: Eastwood has been more focused on directing
and producing in recent years rather than acting. While he has occasionally
taken on acting roles, it's uncertain if he would choose to return to
an acting role, especially one that he originally made famous decades ago.
Creative Direction: Remakes often seek to bring a fresh perspective to the
story, which can involve casting new actors to embody the characters.
While bringing back original actors can add a sense of nostalgia, it's
not always aligned with the creative vision for a new adaptation.
If you're interested in updates regarding this potential project,
I recommend keeping an eye on official announcements from reputable
entertainment news sources or statements from Clint Eastwood and
the production companies involved. Until such information becomes
available, the idea of Clint Eastwood reprising his role in a 2025
remake remains speculative.

通过 Perplexity.ai 比较 o1 和 o1-mini 之间的差异

Source: Author/LN Perplexity.ai

虽然这是一个简单的提示,但正如我之前提到的,X 正在热闹地展示多个高级用例。这条推文突出显示其中之一。

卡帕希谈论模型懒惰

接待与评论

01模型的接受程度是喜忧参半的。

  • 积极反馈:许多用户和分析人士认为o1型号是人工智能推理能力的重要进步,对于发展更自主的人工智能系统至关重要。分析人士指出,这一进步可能导致需要推理作为基本技能的更有能力的人工智能代理人。
  • 顾虑:尽管有一定进步,但仍存在有关o1型号响应速度较慢和成本较高的顾虑。一些专家强调,尽管推理很关键,但并不能解决培训数据中存在的偏见等潜在问题。批评者认为,提高数据质量同样重要,以增强推理模型的效果。
  • 限制:目前的o1型号缺乏先前型号中存在的某些功能,如浏览网页和上传文件,这限制了它们在一些应用中的效用。此外,在处理复杂查询时,它们可能会较慢,有时需要超过十秒才能响应。

結論

OpenAI的o1模型代表了AI推理技术的进步,在处理复杂问题解决任务方面展示了强大的能力。它们在竞技编程和数学推理等领域表现良好,但也需要考虑成本和响应时间。虽然这些模型提供了改进的推理能力,用户和开发人员需要权衡这些益处与集成挑战和费用等实际因素。随着技术的不断发展,看到OpenAI如何解决这些初期限制以及这些模型如何在各种应用中被采用将是很有趣的。

尾件:

我通常请ChatGPT根据1到5的评分标准对我的文章进行评分和审阅。以下是我从“o1预览”收到的“最终”审阅:)

Source: Author/LN o1-preview

参考资料:

  1. 学习用LLMs推理 | OpenAI
  2. OpenAI 据报道正在以 1500 亿美元的估值进行筹款-使其价值超过 Fortune 500 公司中的 88% | Business Insider India

2024-09-15 04:26:00 AI中文站翻译自原文