Name: 大型語言模型簡要說明 (Large Language Models explained briefly)
Uploaded: 2025-06-19T13:51:48.000Z
Duration: 7 min 58 s

Imagine you happen across a short movie script that describes a scene between a person and their AI assistant.

想象一下，你偶然發現了一個電影劇本短片，描述了一個人和他的人工智能助手之間的場景。

The script has what the person asks the AI, but the AI's response has been torn off.

劇本里有這個人向人工智能提出的問題，但人工智能的回答卻被撕掉了。

Suppose you also have this powerful magical machine that can take any text and provide a sensible prediction of what word comes next.

假設你也有一臺功能強大的神奇機器，它可以處理任何文本，併合理預測下一個單詞是什麼。

You could then finish the script by feeding in what you have to the machine, seeing what it would predict to start the AI's answer, and then repeating this over and over with a growing script completing the dialogue.

然後，您可以向機器輸入所需的內容，看它會預測出什麼來開始人工智能的回答，然後重複這樣的過程，用不斷增加的腳本來完成對話。

When you interact with a chatbot, this is exactly what's happening.

當你與哈拉機器人互動時，這正是正在發生的事情。

A large language model is a sophisticated mathematical function that predicts what word comes next for any piece of text.

大型語言模型是一種複雜的數學函數，可預測任何文本的下一個單詞。

Instead of predicting one word with certainty, though, what it does is assign a probability to all possible next words.

不過，它並非以必然性預測單一詞彙，而是為所有可能的後續詞彙分配機率值。

To build a chatbot, what you do is lay out some text that describes an interaction between a user and a hypothetical AI assistant, you add on whatever the user types in as the first part of that interaction, and then you have the model repeatedly predict the next word that such a hypothetical AI assistant would say in response, and that's what's presented to the user.

要構建一個哈拉機器人，你要做的就是鋪設一些文本，描述用戶與假想的人工智能助手之間的互動，你要添加用戶輸入的任何內容作為互動的第一部分，然後讓模型反覆預測假想的人工智能助手作為迴應會說的下一個詞，這就是呈現給用戶的內容。

In doing this, the output tends to look a lot more natural if you allow it to select less likely words along the way at random.

在這樣做的過程中，如果允許它沿途隨機選擇不太可能出現的單詞，輸出結果往往會看起來更自然。

So what this means is even though the model itself is deterministic, a given prompt typically gives a different answer each time it's run.

是以，這意味著即使模型本身是確定的，但每次運行時，給定的提示通常會給出不同的答案。

Models learn how to make these predictions by processing an enormous amount of text typically pulled from the internet.

模型通過處理通常從互聯網上提取的大量文本來學習如何做出這些預測。

For a standard human to read the amount of text that was used to train GPT-3, for example, if they read non-stop 24-7, it would take over 2600 years.

舉例來說，一個標準的人類要閱讀用於訓練 GPT-3 的大量文本，如果他們全天候不停地閱讀，則需要 2600 多年的時間。

Larger models since then train on much, much more.

從那時起，更大型的機型的訓練內容要多得多。

You can think of training a little bit like tuning the dials on a big machine.

你可以把訓練想象成調試一臺大機器上的刻度盤。

The way that a language model behaves is entirely determined by these many different continuous values, usually called parameters or weights.

語言模型的行為方式完全由這些不同的連續值（通常稱為參數或權重）決定。

Changing those parameters will change the probabilities that the model gives for the next word on a given input.

更改這些參數將改變模型根據給定輸入給出下一個詞的概率。

What puts the large in large language model is how they can have hundreds of billions of these parameters.

大型語言模型之所以『大』，在於它們能夠擁有高達數千億的這些參數。

No human ever deliberately sets those parameters.

Instead, they begin at random, meaning the model just outputs gibberish, but they're repeatedly refined based on many example pieces of text.

相反，它們開始時是隨機的，這意味著模型只會輸出胡言亂語，但它們會根據許多文本示例反覆改進。

One of these training examples could be just a handful of words, or it could be thousands, but in either case, the way this works is to pass in all but the last word from that example into the model, and compare the prediction that it makes with the true last word from the example.

其中一個訓練示例可能只有幾個單詞，也可能有數千個單詞，但無論哪種情況，其工作原理都是將該示例中除最後一個單詞之外的所有單詞輸入模型，然後將其預測結果與該示例中真正的最後一個單詞進行比較。

An algorithm called backpropagation is used to tweak all of the parameters in such a way that it makes the model a little more likely to choose the true last word, and a little less likely to choose all the others.

我們使用一種名為反向傳播的算法來調整所有參數，使模型選擇真正的最後一個單詞的可能性更大一些，而選擇所有其他單詞的可能性更小一些。

When you do this for many, many trillions of examples, not only does the model start to give more accurate predictions on the training data, but it also starts to make more reasonable predictions on text that it's never seen before.

當你在數以萬億計的示例中這樣做時，模型不僅能對訓練數據做出更準確的預測，還能對從未見過的文本做出更合理的預測。

Given the huge number of parameters and the enormous amount of training data, the scale of computation involved in training a large language model is mind-boggling.

鑑於參數數量龐大，訓練數據量巨大，訓練一個大型語言模型所需的計算量令人瞠目結舌。

To illustrate, imagine that you could perform 1 billion additions and multiplications every single second.

舉例說明：想象一下，每秒鐘可以進行 10 億次加法和乘法運算。

How long do you think it would take for you to do all of the operations involved in training the largest language models?

您認為需要多長時間才能完成訓練最大語言模型所涉及的所有操作？

The answer is actually much more than that, it's well over 100 million years.

答案其實遠不止這個數字，而是遠遠超過了 1 億年。

This is only part of the story though, this whole process is called pre-training.

但這只是故事的一部分，整個過程被稱為預培訓。

The goal of auto-completing a random passage of text from the internet is very different from the goal of being a good AI assistant.

自動完成一段互聯網上的隨機文本的目標與成為一個優秀的人工智能助手的目標截然不同。

To address this, chatbots undergo another type of training, just as important, called reinforcement learning with human feedback.

為了解決這個問題，哈拉機器人還要接受另一種同樣重要的訓練，即通過人類反饋進行強化學習。

Workers flag unhelpful or problematic predictions, and their corrections further change the model's parameters, making them more likely to give predictions that users prefer.

工作人員會標記無用或有問題的預測，他們的修正會進一步改變模型的參數，使其更有可能給出用戶喜歡的預測。

Looking back at the pre-training though, this staggering amount of computation is only made possible by using special computer chips that are optimized for running many, many operations in parallel, known as GPUs.

不過，回過頭來看預訓練，這種驚人的計算量只有通過使用特殊的計算機芯片（即 GPU）才能實現，這些芯片經過優化，可以並行運行很多很多運算。

However, not all language models can be easily parallelized.

然而，並非所有語言模型都能輕鬆實現並行化。

Prior to 2017, most language models would process text one word at a time, but then a team of researchers at Google introduced a new model known as the transformer.

2017 年之前，大多數語言模型都是一個詞一個詞地處理文本，但後來谷歌的研究團隊推出了一種被稱為轉換器的新模型。

Transformers don't read text from the start to the finish, they soak it all in at once, in parallel.

變形金剛不會從頭到尾閱讀文本，他們會並行地一次讀完。

The very first step inside a transformer, and most other language models for that matter, is to associate each word with a long list of numbers.

轉換器以及大多數其他語言模型的第一步就是將每個單詞與一長串數字聯繫起來。

The reason for this is that the training process only works with continuous values, so you have to somehow encode language using numbers, and each list of numbers may somehow encode the meaning of the corresponding word.

原因是訓練過程只適用於連續值，是以必須用數字對語言進行某種編碼，而每個數字列表都可能以某種方式編碼相應單詞的含義。

What makes transformers unique is their reliance on a special operation known as attention.

變壓器的獨特之處在於它依賴於一種被稱為 "注意力 "的特殊操作。

This operation gives all these lists of numbers a chance to talk to one another, and refine the meanings they encode based on the context around, all done in parallel.

這一操作讓所有這些數字列表有機會互相交流，並根據周圍的語境完善它們所編碼的含義，所有這些操作都是並行進行的。

For example, the numbers encoding the word bank might be changed based on the context surrounding it, to somehow encode the more specific notion of a river bank.

例如，"河岸 "一詞的數字編碼可能會根據周圍的語境發生變化，從而以某種方式編碼出更具體的 "河岸 "概念。

Transformers typically also include a second type of operation, known as a feed-forward neural network, and this gives the model extra capacity to store more patterns about language learned during training.

變壓器通常還包括第二種操作類型，即前饋神經網絡，這就為模型提供了額外的容量，可以存儲更多在訓練過程中學習到的語言模式。

All of this data repeatedly flows through many different iterations of these two fundamental operations, and as it does so, the hope is that each list of numbers is enriched to encode whatever information might be needed to make an accurate prediction of what word follows in the passage.

所有這些數據都在這兩個基本運算的多次重複中反覆流動，在此過程中，我們希望每個數字列表都能得到充實，以編碼任何可能需要的資訊，從而準確預測段落中後面的單詞。

At the end, one final function is performed on the last vector in this sequence, which now has had a chance to be influenced by all the other context from the input text, as well as everything the model learned during training, to produce a prediction of the next word.

最後，對這個序列中的最後一個向量執行最後一個函數，這個向量現在有機會受到輸入文本中所有其他上下文以及模型在訓練過程中學到的所有內容的影響，從而預測出下一個單詞。

Again, the model's prediction looks like a probability for every possible next word.

同樣，模型的預測看起來就像是每個可能的下一個單詞的概率。

Although researchers design the framework for how each of these steps work, it's important to understand that the specific behavior is an emergent phenomenon based on how those hundreds of billions of parameters are tuned during training.

尽管研究人员设计了每个步骤的运行框架，但需要理解的是：具体行为实则是上千亿参数在训练过程中不断调试后，所呈现出的涌现现象。

This makes it incredibly challenging to determine why the model makes the exact predictions that it does.

是以，要確定模型為什麼會做出準確的預測，具有極大的挑戰性。

What you can see is that when you use large-language model predictions to autocomplete a prompt, the words that it generates are uncannily fluent, fascinating, and even useful.

你可以看到，當你使用大型語言模型預測來自動完成提示時，它所生成的單詞會異常流暢、引人入勝，甚至非常有用。

If you're a new viewer and you're curious about more details on how transformers and attention work, boy do I have some material for you.

如果你是一名新觀眾，並且對變形金剛和注意力如何工作的更多細節感到好奇，那麼我有一些材料要給你。

One option is to jump into a series I made about deep learning, where we visualize and motivate the details of attention and all the other steps in a transformer.

一種方法是跳入我製作的深度學習系列，我們將注意力的細節和變壓器中的所有其他步驟可視化並加以激勵。

But also, on my second channel I just posted a talk I gave a couple months ago about this topic for the company TNG in Munich.

另外，在我的第二個頻道上，我剛剛發佈了幾個月前我在慕尼黑 TNG 公司就這一主題發表的演講。

Sometimes I actually prefer the content I make as a casual talk rather than a produced video, but I leave it up to you which one of these feels like the better follow-on.

有時，我其實更喜歡把我製作的內容作為一次隨性的談話，而不是製作成視頻，但我讓你們自己決定哪一個更適合作為後續內容。