Placeholder Image

字幕列表 影片播放

由 AI 自動生成
  • So, I want to talk about a prompt that GPT-4L really struggles with, but our new model O1 preview can do pretty well.

    是以,我想談談 GPT-4L 在處理一個提示符時非常吃力,但我們的新型 O1 預覽版卻能做得很好的問題。

  • And the prompt is simple, it's write a six-line poem about squirrels playing koalas at soccer that meets the following constraints.

    提示也很簡單,就是寫一首關於松鼠踢考拉足球的六行詩,並滿足以下限制條件。

  • In line two, the last word should end with I.

    在第二行中,最後一個詞應以 I 結尾。

  • In line three, the second word begins with U.

    在第三行中,第二個單詞以 U 開頭。

  • In line five, the second to last word is eucalyptus.

    在第五行,倒數第二個詞是桉樹。

  • And in the final line, each word has two syllables.

    在最後一行中,每個詞都有兩個音節。

  • So first we'll try with GPT-4L.

    是以,我們首先嚐試使用 GPT-4L。

  • And we'll see that the answer from GPT-4L meets some of the constraints, but not all of them.

    我們會發現,GPT-4L 的答案符合部分限制條件,但不符合全部限制條件。

  • The reason it's hard for GPT-4L is it has to get it correct on the first try.

    對 GPT-4L 來說,這很難,因為它必須在第一次嘗試時就正確無誤。

  • It can't check that it meets the constraints and then revise the poem.

    它無法檢查是否符合約束條件,然後再修改詩歌。

  • Now let's try the same poem with O1 preview.

    現在,讓我們用 O1 預覽版試試同樣的詩句。

  • And we'll see that differing from GPT-4L, O1 preview starts thinking before giving the final answer.

    我們會發現,與 GPT-4L 不同的是,O1 預覽在給出最終答案之前會開始思考。

  • And you can view a summary of the thinking process of the model.

    您還可以查看該模型的思維過程摘要。

  • So first you can see it's starting to think about different words for rhyming.

    首先,你可以看到它開始思考不同的押韻詞。

  • Then you can see it wants to make sure the last word matches I.

    然後,您可以看到它希望確保最後一個詞與 I 匹配。

  • It thinks about words like alibi.

    它在思考不在場證明之類的詞。

  • It's analyzing word endings, and it's thinking about words like ski.

    它分析詞尾,思考滑雪這樣的詞。

  • Then it's piecing together phrases, but it thinks they don't quite fit.

    然後,它開始拼湊短語,但又覺得不太合適。

  • It's thinking about phrases where the second word starts with U.

    這是在思考第二個單詞以 U 開頭的短語。

  • Then it's tweaking the words to fit the two-syllable rule for line six.

    然後是調整詞語,以符合第六行的雙音節規則。

  • It's digging into various two-syllable word combinations.

    這是挖掘各種雙音節詞的組合。

  • Then it's checking whether the poem aligns with all the guidelines.

    然後是檢查詩歌是否符合所有準則。

  • It's working through the poem to analyze the soccer aspect.

    這是通過詩歌來分析足球方面的內容。

  • And now let's look at the final poem.

    現在讓我們看看最後一首詩。

  • So in the second line, the word safari does end with I.

    是以,在第二行中,"safari "一詞確實以 "I "結尾。

  • In the third line, the second word unleash does begin with U.

    在第三行中,第二個詞 unleash 確實是以 U 開頭的。

  • In the second to last line, eucalyptus, the second to last word is eucalyptus.

    倒數第二行,桉樹,倒數第二個詞是桉樹。

  • And finally, in the final line, under moonlight creature scatter, indeed, each word has two syllables.

    最後,在最後一行 "月光下的生物散落 "中,每個詞確實都有兩個音節。

  • So this is an example of a prompt where, because the model can generate candidates and do reasoning before giving the final answer, it's able to give a higher quality response.

    是以,這是一個提示的例子,由於模型可以生成候選答案,並在給出最終答案前進行推理,是以能夠給出更高質量的答案。

So, I want to talk about a prompt that GPT-4L really struggles with, but our new model O1 preview can do pretty well.

是以,我想談談 GPT-4L 在處理一個提示符時非常吃力,但我們的新型 O1 預覽版卻能做得很好的問題。

字幕與單字
由 AI 自動生成

單字即點即查 點擊單字可以查詢單字解釋