仍在进化的ChatGPT变得更强大了吗?(中)
日期:2023-03-25 12:30

(单词翻译:单击)

Ym]Usa_ejTvGN&5(N[jI~!6QFT1jE

How is GPT-4 different from what we might have experienced if we messed around with, say, ChatGPT a few days before this?

fNcIt)fEOc|mmqLRfe

GPT-4与我们在几天前摆弄ChatGPT时体验到的服务有何不同?

zXpg@|WB|==UbJ*

Yeah, so the big thing -- So, you know, when you look at it, it really just looks like the same old chatbot as before.

+X)xs2tOnTg=)uEo9

是的,最重要的是——当你观察它的外观时,觉得他与之前的聊天机器人别无他样.E(RUCR[zX_5

F[ZhsWdoxo*l,~,*er

You can talk to it. It can respond in all of these elegant ways. You can tell it to, you know, write a poem about a lion in space and it will do that, you know, almost immediately. It'll read really well.

H=dhb08GNvOw]Fc9((m

你可以对它说话,它则会用文雅的语言进行回复sY&SyI_.3^lA。你可以以太空中的狮子为对象让它赋诗一首,它将会立即产出,而且成果也不错W3~~j~l9x|

mL2HJ!Gw|*m

OpenAI, the research lab that has made it, says it's more accurate in what it says, that it's better at sensing, you know, commands that will break the rules and saying no to those.

#zV-I@Bce7I6MUe

制作GPT-4的研发实验室OpenAI表示其语言准确性很高,更擅长感知违反规则的指令,并对这些指令说不5&.!Z5,8#%ZhtShv

JAN^2OsLw),sWk

And it's also better at knowing which requests are actually harmless and being able to respond to those.

FUIAjq]vF,-KKPZ2egO

GPT-4能更好地识别哪些要求无害,并能对这些要求做出回应D&Lf(s|99ihFe@K

hjs#Qi!.-3&AXu[;V@tg

I mean, there are some controls that OpenAI has folded in like, "Don't you know -- Don't use this to do violent things or target people for harassment or do any kinds of like sketchy criminal stuff."

n!r_Gs8W;8-hAX(SF*

OpenAI已经加入了一些控制装置,比如,“这你可知道——不能用它实施暴力,对别人进行骚扰,或者实施任何类似不好的犯罪行为@s~&q*PbN*。”

IT4l5+jZWM=D)W^@

And so GPT-4 is just better at responding to those kinds of things. It's also just more like cogent.

d)|b|[twOh6a

GPT-4能更好地对这些事情做出反应m6)~IVsYKUUT@_]。回答也更合乎逻辑8U(8Cg(oHW-Tp

SSYV-@PdIr

It has been shown by OpenAI to pass all of these, you know, AP and bar exams.

D%PgQxz%nw=jmpQ]GV

OpenAI已经证明GPT-4可以通过所有AP考试与律师考试dM*4twZVaqw%1fX_D%

MtUr6OaU!S

There was a meme of like a bunch of chicken nuggets arranged in the pattern of like what looks like the world from space. And it had this, like, really, like, highfalutin, fancy phrase on top.

Hb*csDK@n*

有这样一张热图:一堆鸡块排列成从太空俯视地球的样子,图片上还附有夸张有趣的话语s_5J~t7Co~@7-u+;

i8AIYs6ztR#

And somebody, you know, posted that meme and was like, "GPT-4, why is this funny?"

,)S9JA~L,_

有人上传了那个热图,问道:“GPT-4,为什么这很有趣?”

ak]V3fIgxQR|u.

And GPT-4 was able to, like, look at this image and assess that well, it's funny because it's like this silly image of chicken nuggets in the world attached to this very, like, emotional, poignant saying.

%tSOxQD6Y4

GPT-4能够观察这张图片并很好地进行评估,它回复这很有趣,因为用鸡块摆成的傻傻的地球图片与一语中的的情绪化语言联系在了一起pXAA5&l62,;F#J*VOw

FOEx4PR=qTa)[IG

So that's, like, really kind of cool because, you know, when you're working with images, it's a whole other form of understanding that the A.I. has to go through.

E[GFV)epm^@sK^)

这真的很厉害,因为当处理图像时,人工智能必须经历另一种完全不同形式的理解;D=mjXsd9mb&M(m

AtCD.azp%J

It's not just, you know, a bunch of words, but it's, like, diagnosing what's in a single image and then being able to, like, talk about it in a elegant way.

U8VhMMY8nNmM

这不仅仅是一堆词语,更像是诊断一张图片上的内容后能够用文雅的语言讨论它+0n,bm=Q6F

8vjRh9WRCfbBpK!a0;%

The image part is, I think, really the most interesting advance, and it's also so advanced that OpenAI has said, "We're a little afraid to share this publicly."

T2#l~]Judclk

我认为图像部分真的是最有趣的进步,这项技术如此先进,以至于OpenAI说,“我们有点害怕公开分享这项技术!0)j8setL_VY1,)。”

E]ZNOw6%IH09pki

Like, "We don't really know how people will be using it. So it's only really allowed for specific researchers right now. And we may open it to the public when we feel more comfortable."

5z=7ih&0!gm+B

“我们真的不知道人们会如何使用这项技术KY0MRj[QucL1h8D,gti。所以,目前只允许特定的研究人员使用VwRnot3pr_c_R。当我们感到舒服的时候,可能会向公众开放EoFu-)HG[YNt。”

%X=0zs~WRaZ|CR0uoC|zhJ4OP*cd;kCh#DmsH],u4b
分享到