Certainly! Here's the translation in simplified Chinese, maintaining the HTML structure: ```html LLM,三思而后言! ``` This HTML snippet will display "LLM,三思而后言!" in simplified Chinese.

Sure, here's how you can translate "Why ChatGPT and its companions could use some thoughtful reflection" into simplified Chinese while keeping the HTML structure: ```html 为什么ChatGPT及其伙伴们需要进行一些深思熟虑的反思 ``` In the HTML snippet above: - `` tags are used to encapsulate the translated text. - The Chinese translation "为什么ChatGPT及其伙伴们需要进行一些深思熟虑的反思" accurately reflects the meaning of the original English sentence. This maintains the structure of HTML while providing the translated content in simplified Chinese.

A robot looking into an infinity mirror, with a thoughtbubble containing an iteratively refined idea.

Sure, here is the HTML structure with the translated text in simplified Chinese: ```html

在这篇博客中,我将分享我对大型语言模型(LLMs)进展放缓的看法,以及为什么当前的自回归实现对于复杂推理不足够。然后,我将讨论我们如何避免这个问题,学习人类思维过程。最后,我会论述这将导致在自回归模型之上更偏爱扩散模型。这种迭代思维的实现可能是通向人工推理的关键步骤。

``` In this HTML snippet: - `

` tags are used to denote paragraphs, suitable for the translated text. - The translated text is placed directly within the `

` tags in simplified Chinese. This ensures the translated content maintains the structure and format suitable for HTML, while accurately reflecting the intended message in Chinese.

Sure, here's how you would write "The current state of LLMs" in simplified Chinese while keeping the HTML structure intact: ```html LLMs 的当前状态 ```

```html

自回归大语言模型(LLM)如ChatGPT在文本生成方面取得了显著进展,能够生成连贯且与语境相关的内容。利用这种范式进行文本生成已经在自然语言处理方面取得了显著的进展。然而,尽管它们取得了令人瞩目的成就,LLM的改进开始放缓,这表明它们可能正接近其能力的上限。这些能力的限制部分原因是我们正在耗尽新的训练数据来源,计算能力的增长相对缓慢,以及这些资源的回报递减。然而,当前LLM的一些缺陷似乎与其自回归实现方式有关,我认为通过转向不同的生成范式可能更容易解决这些问题。

```

```html

自回归模型及其缺点需要注意的是,对于本故事来说,“自回归”在LLM(语言模型)的上下文中指的是一种特定类型的(文本)生成方法。该方法通过预测序列中的下一个词来生成文本,其预测依据是该序列中的先前单词。这也是为什么LLM也被称为“下一个词预测器”的原因。该过程的示例如图1所示。

```

Certainly! Here's the translated text in simplified Chinese, while keeping the HTML structure intact: ```html

我认为当前语言模型的一些主要缺点是由于它们依赖于这种顺序数据处理方式造成的。这导致了多种问题,这些问题固有于自回归处理中,无法通过增加训练数据集的大小、模型的规模或训练的长度来克服。

``` In simplified Chinese characters: ``` 我认为当前语言模型的一些主要缺点是由于它们依赖于这种顺序数据处理方式造成的。这导致了多种问题,这些问题固有于自回归处理中,无法通过增加训练数据集的大小、模型的规模或训练的长度来克服。 ``` This translation should fit within an HTML `

` paragraph element.

Figure 1: Illustration of the autoregressive method’s output generation process. This method produces text sequentially, predicting each word based on the previous words in the sequence.

```html

顺序依赖性问题是生成开始时的词汇对最终输出具有决定性影响,并且一旦生成后就无法更改。这会降低复杂输出的质量,因为语言模型没有机会思考自己的输出,然后必要时予以撤回。在单个回答中,语言模型很少会改变其观点,因为这在(书面)训练数据中也很少见。只有在特别查询的情况下,语言模型才会检查其陈述是否正确。这通常被描述为“思维链”提示,但这需要通过向语言模型额外查询来进行人为干预。

```

```html

随机性质与第一个问题相关联的第二个问题是选择下一个“最有可能”词语的随机性质。因为我们通常给模型一个特定的“温度”参数,允许它选择最有可能的前 X 个词之一,所以总是有可能在回答的早期做出次优决策。在接下来的回答过程中,模型将不得不处理这个次优选择,从而加重它最初的错误。在过程的早期做出的这个单一次优词选择可能导致一系列尴尬的、不正确的或不连贯的想法。特别是对于更难的问题,可能有两个答案都很有可能但只有一个正确,这可能会带来问题。如果一个语言模型对逻辑推理问题的结果有70%的确信度,对于A或B答案,它仍然可能只有70%的时间回答正确(取决于“温度”参数的实现方式)。

```

```html

缺乏远见的第三个问题是,连续生成的模型不知道在序列的末尾将生成什么内容,而这对某些问题实际上是必要的。一个非常明显的例子是这样的问题:“你的下一句将包含多少个词?”。如果允许你先写下一句,然后再纠正最初猜测的数量,回答这个问题就容易得多。这似乎是一个显而易见的技巧,但实际上在现实生活中经常发生。另一个出现这个问题的例子是,当你让一个语言生成模型写一篇论述文,讨论它是支持还是反对某个问题。如果不先写下论文并列出正反两方面的观点,可能很难在一开始就想出一个有说服力的标题。

```

Certainly! Here's the simplified Chinese translation of the text, keeping the HTML structure: ```html

当前语言模型(LLMs)存在许多其他缺陷:如无法持续学习、易产生幻觉、固有的偏见等等。然而,上述三个问题与当前语言模型的自回归特性直接相关。我认为通过稍微不同的网络架构实现,这些问题是最容易解决的,因此是低 hanging fruit。

``` This HTML snippet contains the translated text in simplified Chinese, maintaining the structure for integration into an HTML document.

Certainly! Here's the translation of "The similarities with the human thought process" in simplified Chinese, while keeping the HTML structure: ```html

人类思维过程的相似之处

```

Certainly! Here's the HTML structure with the translated text in simplified Chinese: ```html

在我们继续处理这些问题的潜在解决方案之前,我首先想考虑语言模型(LLMs)在‘思考’方式上与人类的(不)相似之处。通过这一点,我们可以看到人类思维如何解决这个问题,然后开始在我们的模型中实施这一点。

``` This HTML snippet preserves the structure while including the translated text in simplified Chinese.

Certainly! Here's the translated text in simplified Chinese, keeping the HTML structure intact: ```html

免责声明:尽管本文简要讨论了哲学概念,如人工智能‘思维’的本质以及意识与潜意识之间的区别,但并不旨在对这些问题提供明确答案。对于某些有争议的术语的使用是在隐喻意义上,用来阐述思想,而不是精确定义生命中这些复杂问题的答案。

``` This HTML snippet preserves the structure while providing the simplified Chinese translation of the English text you provided.

```html

有趣的是,自回归语言模型“思考”的方式反映了人类思维过程的某些方面。我们赋予语言模型在某种程度上能够以介于“意识流”和快速对话之间的方式思考,词语流动几乎与思维同步。类似地,语言模型也会像“思维快速与慢”的作者丹尼尔·卡尼曼在其著作中所描述的那样,仅仅产生下一个“脑海中浮现”的词语(或标记)。这种思维方式通常也被称为“系统1”思维。

``` Here is the HTML structure with the text translated into simplified Chinese.

```html

这种生成方式对人类的头脑风暴非常有用,因为在这里,想法可以失败、得以改进,并逐步改善。然而,这种方法并不适合更为深思熟虑和连贯的写作过程,人们在写作之前通常会先思考,或者至少有机会进行重写。

```

Certainly! Here is the translated text in simplified Chinese, keeping the HTML structure intact: ```html

一个以纯粹的“意识流”方式写成的人类文本的著名例子是詹姆斯·乔伊斯的《尤利西斯》*。我认为许多(不富经验的)读者会同意,这并不是我们希望从我们的AI模型中得到的输出。相反地,我认为许多研究人员会同意,训练于数据过少、参数过少的较小AI模型的输出,读起来与那本书中的文本非常相似。以下方框展示了两者的例子。

``` This HTML snippet contains the translated text in simplified Chinese while maintaining the original structure for web display.

Sure, here is the translation in simplified Chinese while keeping the HTML structure: ```html

作者并未声称阅读完整本《尤利西斯》,也不意图侮辱任何更加文学倾向的读者。

``` In this HTML snippet: - `

` denotes a paragraph in HTML. - `作者并未声称阅读完整本《尤利西斯》,也不意图侮辱任何更加文学倾向的读者。` is the translated text in simplified Chinese. This HTML structure ensures the content is presented as a paragraph, typical for web page formatting.

Ulysses by James Joyce[1]:

Bare clean closestools waiting in the window of William Miller, plumber,
turned back his thoughts. They could: and watch it all the way down,
swallow a pin sometimes come out of the ribs years after, tour round
the body changing biliary duct spleen squirting liver gastric juice
coils of intestines like pipes. But the poor buffer would have to stand
all the time with his insides entrails on show. Science.

GPT-2 Small (example produced by Aaron et al.[2]):

“While I was nine-year old Angela, there were 15 of my who came back in
on-the-job and my best shot at life,” she says. “Tiny took away substances
in life, and your mother’s life was financed by a small little gun we just
bashed, and sometimes, I’d end up arguing in the closet with my mother
where she killed her little crow [the tiny squirrel-nay] but couldn’t catch
anything.”

```html

这种半连贯的词汇沙拉不一定会成为问题,因为这就是大多数人相互交谈的方式。从人与人之间来看,如果有人说了一件事,犯了错误,并在下一句中纠正自己,我们是可以接受的。然而,这种与大语言模型(LLM)交流的方式仍然不令人满意,有两个原因。首先,我们习惯了计算机总是提供精确的事实性输出,很难接受我们已经发明了一种如此逼真地模仿真实人类对话的技术。其次,LLM通常生成书面文本。在书面形式上,我们通常期望不正确的答案已经被过滤掉。

```

Certainly! Here's your text translated into simplified Chinese, while keeping the HTML structure: ```html

在面对生成合理、可信(书面)输出的任务时,人类通常不会仅仅说出心中第一个想法。相反,这需要努力、注意力、反思和推理。这正是丹尼尔·卡内曼所称的“系统2”思维。

``` In Chinese: ```html

在面對生成合理、可信(書面)輸出的任務時,人類通常不會僅僅說出心中第一個想法。相反,這需要努力、注意力、反思和推理。這正是丹尼爾·卡內曼所稱的“系統2”思維。

``` This HTML snippet preserves the structure and presents the translated text in simplified Chinese.

Sure, here's the translated text in simplified Chinese, while keeping the HTML structure intact: ```html

有多种方法可以在向世界传播之前组织和打磨你的想法。那么,我们如何解决从无序到有序思维的问题呢?我们能否在语言生成模型的实现中模仿这个解决方案呢?

``` In this HTML snippet: - `

` denotes a paragraph tag, used to structure the text. - The Chinese text provided is the translation of the original English text.

Certainly! Here's how you can structure your HTML to include the translated text in simplified Chinese: ```html

Translated Text

我将人类采用的解决方案分为两类,因为我将使用它们来等同于我们改进语言模型的不同方法。我将这些方法称为“外部精炼”和“内部精炼”。事实上,根据具体情况,我们(人类)经常在不同程度上同时采用内部和外部的方法。

``` In the translated HTML: - `` specifies that the language is simplified Chinese. - The text inside the `

` tag is the translated content. This HTML structure will display the translated text in a web browser while maintaining the proper language encoding and structure.

Certainly! Here's the simplified Chinese translation of the text while keeping the HTML structure intact: ```html

通过首先向世界表达您的思想草稿,几乎像意识流一样,例如在撰写初稿或进行非正式对话时,可以对思想进行外在的精炼。然后通过重新排序、删除、添加和重写来实现思想的精炼。在对话中,通过互相纠正并基于论点重新考虑您的观点,也可以实现思想的精炼。这个过程可以一直持续,直到形成一种连贯、合理的思想。

``` This HTML snippet preserves the structure and includes the simplified Chinese translation of the provided text.

Sure, here is the translated text in simplified Chinese, maintaining the HTML structure: ```html

思维的内在完善是一个认知过程,涵盖头脑中概念的构思、评估和反复完善,而无需明确的口头表达。这一过程还可能包括视觉、听觉和感官元素,因为思想并不仅限于文字。此外,将思想转化为书面形式有时会减少其复杂性,因为一个想法的本质通常难以完全用语言捕捉。内在完善在意识和潜意识层面上同时运作,使思维得以在时间推移中无需明确努力而发展,就像一个人在一个好觉后可能会注意到的那样。

``` This HTML structure preserves the original content while providing the translated text in simplified Chinese.

Certainly! Here's the HTML structure with the translated text in simplified Chinese: ```html

这两种方法的好处在于能够观察和修订完整的思路,这种思路可以等同于LLM的最终输出。LLM通常能够通过其注意力结构查看完整的输入,但只能看到它们在那一点上已经生成的输出。它们无法对生成后的输出进行任何更改,除了追加内容外,因此无法对输出进行迭代改进。

``` In simplified Chinese: ``` 这两种方法的好处在于能够观察和修订完整的思路,这种思路可以等同于LLM的最终输出。LLM通常能够通过其注意力结构查看完整的输入,但只能看到它们在那一点上已经生成的输出。它们无法对生成后的输出进行任何更改,除了追加内容外,因此无法对输出进行迭代改进。 ``` This HTML snippet will display the translated text in a structured paragraph format on a webpage.

Sure, here's the translation of the text into simplified Chinese while keeping the HTML structure intact: ```html

精炼与推理AI面临的最后一大挑战之一是其在“推理”方面的困难。根据剑桥词典的定义,推理被定义为:

``` This HTML snippet provides the translated text in simplified Chinese, maintaining the structure for easy integration into web content.

Sure, here's the HTML structure with the translated text in simplified Chinese: ```html

思考某事以做出决定的过程

``` In simplified Chinese, "The process of thinking about something in order to make a decision" translates to "思考某事以做出决定的过程".

Sure, here's the translated text in simplified Chinese, while keeping the HTML structure intact: ```html

很难定义思维发生的确切地点和时刻,无论是在人类还是AI模型中。然而,通过执行细化步骤,我们至少可以延伸思维可能发生的过程。显然,这种细化通常会提高人类的推理能力。更多的思考时间通常意味着更敏锐的洞察力。在接下来的部分,我将讨论这种通过细化改进的推理能力是否也会转移到AI身上。

``` This HTML snippet contains the translated text in simplified Chinese, structured as a paragraph (`

` tag) for web display.

Certainly! Here's the translation of "Transferring the human refinement process to the machine" in simplified Chinese, while keeping the HTML structure: ```html 将人类的精炼过程转移到机器上 ```

Certainly! Here's the translated text in simplified Chinese while keeping the HTML structure intact: ```html

通过考虑人类的思维过程,我们可以推断出将思维的精炼转移到LLM中的方法。我将讨论如何在LLM的实现中实施外部和内部的精炼过程。

``` In this HTML snippet: - `

` tags denote a paragraph in HTML. - The Chinese text translates to: "通过考虑人类的思维过程,我们可以推断出将思维的精炼转移到LLM中的方法。我将讨论如何在LLM的实现中实施外部和内部的精炼过程。" This translation preserves the structure and meaning of the original English text in simplified Chinese.

```html

LLMs 的思维外部细化已经通过文献中经常讨论的“思维链”实现在应用中。然而,这一过程缓慢且资源密集,因为LLM必须先自回归地形成完整的答案。在这个过程的每个完整迭代中,需要对模型进行多次调用,因为每个标记都需要一个新的函数调用。此外,自回归方法的缺点仍然存在于生成的每个新输出中。

```

To translate the given English text into simplified Chinese while keeping the HTML structure intact, you can use the following: ```html

与其经过这种冗长的外部完善过程,我们更希望在语言模型有机会完善答案之前不要求其写出完整答案。我们宁愿让语言模型像人类拥有的内部完善机制一样,逐步完善答案。为了做到这一点,我们将不得不从自回归生成转变为一种生成范式,使语言模型能够在输出答案给用户之前逐步改进答案。此外,最好是总生成时间不会增加太多。

``` This HTML snippet will display the translated text in simplified Chinese within a paragraph (`

` tag), maintaining the original HTML structure.

```html

在LLMs中,思想的内部精炼可以通过两种方式实现。然而,这两种方法都基于生成AI中已经广泛使用的一种技术。这种方法称为扩散。在描述扩散模型如何模仿人类内部精炼过程之前,我将首先简要讨论这些模型的定义。

```

Certainly! Here's the translated text in simplified Chinese while keeping the HTML structure intact: ```html

关于扩散模型的简短说明

使用扩散模型时,过程始于所需输出的嘈杂版本。这种噪音代表初始的、未经改进的想法。然后模型重复去噪输出,逐步在每一步改进它。这种方法在生成式人工智能中被广泛用于生成图像。

``` This HTML snippet maintains the structure while presenting the translated text in simplified Chinese.

Sure, here's your text translated into simplified Chinese while maintaining the HTML structure: ```html

有一段时间,人们认为自回归网络更适合离散量化的标记,比如词语,而扩散网络更适合连续信号,比如像素和声音。然而,各种研究表明,文本的扩散生成也是可能的。一个小差异是,使用文本扩散时,过程通常从‘空白’标记开始,而不是从所需输出的嘈杂版本开始,尽管存在各种不同的实现方式。

``` This HTML snippet contains your text translated into simplified Chinese.

```html

最后,重要的是要注意,采用扩散并不意味着我们不再使用变压器网络。这不是扩散(一种生成范式)与变压器(一种网络架构)之间的对立。相反,这是从自回归生成到扩散生成的转变。

```

To translate the provided English text to simplified Chinese while maintaining the HTML structure, you can use the following: ```html

显式扩散我将讨论的第一种扩散方法有时被称为显式扩散。它可以被看作是人类内部和外部细化的结合体,因为它仍然需要一些中间的书面步骤。然而,它可能比自回归语言模型使用的思维链方法快得多。该过程的示意如图2所示。如图所示,在第一次迭代之后已经创建了输出的初步草稿,这仅需要进行一次网络评估。然后,在每个步骤之后进一步完善草稿,直到达到预定义的步骤数。

``` This HTML snippet maintains the paragraph structure and translates the English text into simplified Chinese, with the text enclosed in `

` tags for proper formatting within an HTML document.

Figure 2: Illustration of the iterative refinement process in explicit diffusion models. Instead of generating text sequentially like the autoregressive model, the model refines its all output in each step.

Sure, here's the translated text in simplified Chinese, while keeping the HTML structure intact: ```html 这种方法相比自回归方法具有四个优势: ``` This HTML structure ensures the translated text retains its formatting and can be easily integrated into a web page or document.

  • ```html

    首先,最终输出不依赖于最初的几个词,这可以避免初期错误的累积。扩散过程的迭代特性可以过滤掉这些早期的错误,从而解决了自回归生成中存在的“顺序依赖”问题。

    ```
  • Sure, here's the translation in simplified Chinese while maintaining the HTML structure: ```html

    其次,它对于最初由随机确定的猜测的依赖较小。初始猜测并非一成不变,低概率的猜测在多次迭代中更有可能被纠正,从而使模型能够改进和修正其最初的输出。

    ``` In this translation: - "Second," is translated as "其次," which is commonly used to indicate the second point in a list or discussion. - The text is translated into simplified Chinese, maintaining clarity and fidelity to the original meaning. - The structure `

    ` and `

    ` denote the beginning and end of a paragraph in HTML, ensuring proper formatting if the text were to be used in a web context.
  • Certainly! Here's the translated text in simplified Chinese, keeping the HTML structure: ```html

    第三,它可以在每次迭代中考虑完整的输出,从开始到结束,而不仅仅是在回答开始时生成的输出,从而解决了“缺乏远见”的问题。

    ``` In this HTML snippet: - `

    ` denotes a paragraph tag in HTML, used here to encapsulate the translated text for proper structure and semantics. - The Chinese text inside `

    ` tags is the translation of the provided English sentence.

  • ```html Translated Text

    最后,通常需要较少的网络评估。不必为响应中的每个单词评估网络,它可以在一次迭代中生成完整的答案。答案的质量取决于迭代次数,这是一个可调参数。在大多数应用中,这个数字会比生成的总单词数要低。

    ```

Certainly! Here's the translation of your text into simplified Chinese, while keeping the HTML structure intact: ```html

潜在扩散第二种方法是我将要讨论的。它不需要明确表达中介步骤,而是迭代地改进输出的内部(或潜在)表示。这更类似于人类的内部精炼过程,其中一个想法可以在最终阶段之前进行精炼而不产生任何输出。这个过程在图3中有所说明。再次强调,有多种实现方式。

``` In this HTML snippet: - `

` represents a paragraph tag, which is commonly used to structure text content.

Figure 3: Illustration of the iterative refinement process in latent diffusion models. Instead of generating text each iteration, the model refines its internal hidden (latent) state.

Sure, here's the HTML structure with the translated text in simplified Chinese: ```html

它提供了与显式扩散讨论的相同好处,但又有两个额外的好处:

``` In this structure: - `

` denotes a paragraph tag in HTML. - The Chinese text inside `

` is the translation of "It offers the same benefits as discussed for explicit diffusion, but has two added benefits:".

  • Certainly! Here's the HTML structure with the translated text in simplified Chinese: ```html

    首先,类似于人类内在的精炼过程,在每一步中丢失的信息较少。当从潜空间转换为文本输出时,由于自然语言的固有限制,某些信息不可避免地会丢失。通过在最终步骤之前保持和精炼潜在表示,潜在扩散减少了这些损失,保留了生成内容的更多细微差别和微妙之处。

    ``` In simplified Chinese: ```html

    首先,类似于人类内在的精炼过程,在每一步中丢失的信息较少。当从潜空间转换为文本输出时,由于自然语言的固有限制,某些信息不可避免地会丢失。通过在最终步骤之前保持和精炼潜在表示,潜在扩散减少了这些损失,保留了生成内容的更多细微差别和微妙之处。

    ``` This HTML snippet keeps the structure while presenting the translated text in simplified Chinese.
  • Sure, here's the translated text in simplified Chinese, keeping the HTML structure intact: ```html

    其次,由于在生成过程的最后阶段只需一次从潜空间到文本输出的最终解码,因此它提高了网络效率。注意:对于图像生成,这种效率提升要比文本生成大得多。这是因为图像潜在表示通常比图像显式表示(像素)要小。在训练过程中特别是在节省大量内存和处理能力时,避免显式状态非常重要。然而,对于文本生成,潜在状态通常不比显式状态小,因此这种额外的好处可能不那么显著。

    ``` This HTML snippet contains the translated text in simplified Chinese within a paragraph (`

    ` tag).

Certainly! Here is the translated text in simplified Chinese, while keeping the HTML structure: ```html

在这个故事的前面,我曾经提到人类思维的精炼是改进推理的关键因素。因此,如果我们能够通过扩散模型高效地对人工智能进行精炼,我们可以想象这可能也会改善人工智能的推理能力。当然,并不保证这一方法一定有效。虽然神经网络在某些方面模仿了人脑,但真实神经网络与人工神经网络之间仍存在许多差异。

``` This HTML snippet preserves the structure of the text while providing the simplified Chinese translation.

Certainly! Here's the translated text in simplified Chinese, while keeping the HTML structure intact: ```html

那么,这只是空泛的猜测吗?不!叶等人的最新研究[5]表明,在扩散模型中,推理能力得到了显著提升,甚至比那些体积大4.6倍的自回归模型表现更好。他们的扩散模型实现了比经典的思维链生成快27倍的速度提升。此外,他们还表明,预训练的自回归模型可以进行扩散调整。这减少了从头开始训练大模型的需求,促进了从自回归到扩散文本生成的平稳过渡。

``` This translation maintains the original structure and meaning of the English text.

Sure, here is the translated text in simplified Chinese, formatted to be inserted into an HTML structure: ```html

叶等人还显示扩散模型的一个最终附加优势:如果需要更深思考,增加扩散步骤非常容易。研究表明,增加扩散步骤实际上改善了结果。因此,这可以让这种模型的用户告诉模型,在面对困难问题时是否应该深思熟虑,或者如果问题不太复杂,是否只给出第一个猜测。

``` This HTML snippet contains the translated text in simplified Chinese, maintaining the original structure for easy integration into a webpage or document.

Sure, here's how you could structure it in HTML and provide the translation into simplified Chinese: ```html Current development

Current development

当前发展

``` In the code above: - `

Current development

` represents the English text. - `

当前发展

` represents the simplified Chinese translation. Make sure to adjust the charset and language attributes (`lang="zh-Hans"`) appropriately if needed for your specific use case.

Sure, here's the translation of the text into simplified Chinese, while keeping the HTML structure intact: ```html

迄今为止,与众多自回归模型相比,文本扩散模型的实现仅有少数几种。对潜在文本扩散模型的研究甚至更为稀少,像 Llama、ChatGPT、Gemini 或 Claude 等较大的模型尚未采用扩散范式。尽管像苹果公司的研究者们(张等人 [4])展示了令人鼓舞的结果,但扩散模型并未获得显著的关注。可能有几个因素影响了这一点:

``` This HTML snippet translates the provided English text into simplified Chinese.
  • ```html

    Autoregressive models like GPT-4 have set high benchmarks for text generation, leading to a strong preference and investment in these models by the research community and industry. Only a substantial slow-down in the performance increase of autoregressive models might trigger the necessary paradigm shift.

    ``` Translated to Simplified Chinese: ```html

    像GPT-4这样的自回归模型为文本生成设定了高标准,导致研究界和工业界对这些模型表现出强烈的偏爱和投资。只有自回归模型性能提升的显著放缓可能会引发必要的范式转变。

    ```
  • Sure, here's the translated text in simplified Chinese while maintaining the HTML structure: ```html

    大量的出版物、时间和资源已被投入到改进和调整自回归模型中,从而实现了高度优化的性能。用一种新的、不太成熟的技术超越这些经过精细调整的模型是具有挑战性的,因为在扩散模型中仍有许多改进空间。

    ``` This HTML snippet contains the translated text: "大量的出版物、时间和资源已被投入到改进和调整自回归模型中,从而实现了高度优化的性能。用一种新的、不太成熟的技术超越这些经过精细调整的模型是具有挑战性的,因为在扩散模型中仍有许多改进空间。"
  • Certainly! Here's the translation of the provided English text into simplified Chinese, while keeping the HTML structure intact: ```html

    支持扩散模型的生态系统,包括软件库、预训练模型和开发工具,并不像自回归模型那样成熟。这阻碍了实验和采用。

    ``` This HTML snippet now contains the text translated to simplified Chinese.
  • ```html

    最后,尽管扩散模型在理论上具有优势,但它们实际上可能并不比自回归模型表现更好。这可能导致了对扩散模型研究数量上的幸存者偏差,因为表现不佳的模型往往不会被报道。然而,最近几个月来,来自谷歌Deepmind [6] 和康奈尔大学 [7] 等知名机构的对扩散模型的研究有所增加,这表明扩散模型可能正在受到关注。

    ```

```html

人工智能研究的前景
我预测当前语言模型性能提升速度放缓将导致对新型模型架构和生成范式的热情重新高涨。虽然扩散模型有望修复当前人工智能面临的一些固有问题,但仍存在许多其他问题,并且还有许多不同的实现方式。对于较小的研究团队来说,通常无法与大公司生产的大模型竞争,因为他们缺乏资源。因此,我主张较小的研究团队专注于实施原创解决方案,因为这类研究能够获得比现有架构的渐进增益更多的价值。

``` This HTML structure contains the translated text in simplified Chinese.

Certainly! Here's the HTML structure with the translated simplified Chinese text: ```html

最后,我很想听听您的意见。请告诉我您是否同意或不同意我的观点,如果您对LLM有其他(更好的?)想法,或者您有兴趣在这个主题上合作!

``` In simplified Chinese: ```html

最后,我很想听听您的意见。请告诉我您是否同意或不同意我的观点,如果您对LLM有其他(更好的?)想法,或者您有兴趣在这个主题上合作!

``` This HTML `

` tag will display the translated text in a paragraph structure on a web page.

Certainly! Here's how you can write "References" in simplified Chinese within an HTML structure: ```html 参考文献 ``` In this example, `` is used to maintain the HTML structure, surrounding the translated term "参考文献" which means "References" in simplified Chinese.

Certainly! Here's how you can write "Joyce J. Ulysses. 1922." in simplified Chinese within an HTML structure: ```html

Joyce J. 尤利西斯。1922年。

``` This HTML snippet maintains the structure while presenting the text in simplified Chinese characters.

Sure, here's the translation in simplified Chinese while keeping the HTML structure intact: ```html

2. Lou A, Meng C, Ermon S. 通过估计数据分布的比值进行离散扩散建模. arXiv [stat.ML]. 2023. 可访问: http://arxiv.org/abs/2310.16834

``` In simplified Chinese: ```html

2. Lou A, Meng C, Ermon S. 通过估计数据分布的比值进行离散扩散建模. arXiv [stat.ML]. 2023. 可访问: http://arxiv.org/abs/2310.16834

``` This maintains the structure of an HTML paragraph (`

` tag) and includes the Chinese translation of the citation text within it.

To translate the given citation to simplified Chinese within an HTML structure, you would typically use the following format: ```html

3. Lovelace J, Kishore V, Wan C, Shekhtman E, Weinberger KQ. Latent Diffusion for Language Generation. arXiv [cs.CL]. 2022. Available: http://arxiv.org/abs/2212.09462

``` Now, let's provide the translated text in simplified Chinese: ```html

3. Lovelace J, Kishore V, Wan C, Shekhtman E, Weinberger KQ. 潜在扩散用于语言生成. arXiv [cs.CL]. 2022. 可获取: http://arxiv.org/abs/2212.09462

``` This HTML structure ensures the citation remains clear and the link is preserved.

Certainly! Here's how you can structure your HTML to include the citation in simplified Chinese: ```html

4. Zhang Y, Gu J, Wu Z, Zhai S, Susskind J, Jaitly N. PLANNER: Generating Diversified Paragraph via Latent Language Diffusion Model. arXiv [cs.CL]. 2023. Available: http://arxiv.org/abs/2306.02531

``` Translated to simplified Chinese: ```html

4. 张瑜,顾杰,吴志,翟硕,Susskind J,Jaitly N. PLANNER: 通过潜在语言扩散模型生成多样化段落. arXiv [cs.CL]. 2023. 可获取: http://arxiv.org/abs/2306.02531

``` Make sure to replace the English text with the Chinese translation within the `

` tag while keeping the structure intact.

To translate the given English text to simplified Chinese while keeping the HTML structure intact, you can use the following code snippet: ```html

5. Ye J, Gong S, Chen L, Zheng L, Gao J, Shi H, et al. Diffusion of Thoughts: Chain-of-Thought Reasoning in Diffusion Language Models. arXiv [cs.CL]. 2024. Available: http://arxiv.org/abs/2402.07754

``` In simplified Chinese, it would look like this: ```html

5. 叶杰,龚松,陈雷,郑磊,高健,史浩 等. 思维的传播:扩散语言模型中的思维链推理. arXiv [cs.CL]. 2024. 可访问: http://arxiv.org/abs/2402.07754

``` This code maintains the structure and format of the original English citation while providing the simplified Chinese translation.

Certainly! Here's how you can structure your HTML while displaying the citation in simplified Chinese: ```html

6. Shi J, Han K, Wang Z, Doucet A, Titsias MK. Simplified and Generalized Masked Diffusion for Discrete Data. arXiv [cs.LG]. 2024. Available: http://arxiv.org/abs/2406.04329

``` And the translated text in simplified Chinese: ```html

6. Shi J, Han K, Wang Z, Doucet A, Titsias MK. 离散数据的简化和泛化蒙版扩散. arXiv [cs.LG]. 2024. 可获取:http://arxiv.org/abs/2406.04329

``` This HTML structure will maintain the original citation format while displaying the translated text in simplified Chinese.

Certainly! Here's the translated HTML structure with the simplified Chinese text: ```html

7. Sahoo SS, Arriola M, Schiff Y, Gokaslan A, Marroquin E, Chiu JT, 等. Simple and effective masked diffusion language models. arXiv [cs.CL]. 2024. Available: http://arxiv.org/abs/2406.07524

``` In simplified Chinese: ```html

7. Sahoo SS, Arriola M, Schiff Y, Gokaslan A, Marroquin E, Chiu JT, 等。简单而有效的蒙版扩散语言模型。arXiv [cs.CL]。2024。可访问:http://arxiv.org/abs/2406.07524

```

2024-07-23 04:26:13 AI中文站翻译自原文