GPT-4o Mini First Impressions: Fast, Cheap, & Dang Good.

Summary

OpenAI发布了新的GPT 40 Mini模型，旨在替代GPT 3.5，成为成本效益更高的小型模型。该模型在多语言处理和图像识别方面表现出色，支持并行多模型调用和快速处理大量文本。尽管其性能不及GPT 4 Omni，但价格更低廉，适用于广泛的AI应用。同时，GPT 40 Mini是首个应用新指令层级方法的模型，提高了安全性和可靠性。

Takeaways

• 📈 7月15日，Open AI发布了一个名为GPT 40 Mini的新模型，这是一个新的公共模型。
• 🚀 GPT 40 Mini旨在成为成本效益最高的小型模型，用以替代GPT 3.5，并为免费版Chat GPT提供支持。
• 💰 GPT 40 Mini的输入令牌价格为每百万15美分，输出令牌价格为每百万60美分，比之前的前沿模型便宜60%。
• 🏆 GPT 40 Mini在MLU（机器语言理解）测试中得分82%，表现优于原始的GPT 4。
• 🔍 该模型支持并行多模型调用、处理大量上下文、代码库对话历史或与客户支持互动等用例。
• 🌐 GPT 40 Mini还支持视觉功能，未来还将支持音频输入和输出。
• 📚 GPT 40 Mini是第一个应用新指令层次结构方法的模型，这有助于提高模型抵抗越狱、提示注入和系统提示提取的能力。
• 📅 关于其他模型和功能，Open AI在聊天GPT应用中提供了更新，预计在7月下旬开始对高级语音模式进行Alpha测试。
• 🔮 预计GPT 5可能不会在今年发布，而是可能在明年3月左右发布。
• 🌟 Sora是Open AI的另一个项目，最近在YouTube上发布了更多相关内容，预计今年晚些时候可能会有公开发布。
• 📊 GPT 40 Mini在图像识别和复杂问题处理方面表现出色，尽管与GPT 4 Omni相比仍有差距，但性价比很高。

Q & A

什么是GPT 40 Mini？

GPT 40 Mini是Open AI发布的一种新型的公共模型。它是一种成本效益高的小型模型，旨在取代GPT 3.5，并为不需要GPT 4 Omni或GPT 4 Turbo等高级智能水平的应用提供支持。

GPT 40 Mini的主要特点是什么？

GPT 40 Mini的主要特点是成本低廉、速度快。它在MLU（机器语言理解）测试中得分为82%，表现优异，并且在某些基准测试中超过了原始的GPT 4。此外，它还支持多模态功能，如视觉和音频输入输出。

GPT 40 Mini的定价如何？

GPT 40 Mini的定价非常低廉，每百万输入令牌仅15美分，每百万输出令牌60美分，比之前的前沿模型便宜60%，也比GPT 3.5 Turbo便宜。

GPT 40 Mini在哪些应用场景中表现良好？

GPT 40 Mini适用于需要并行多个模型调用、处理大量上下文、代码库对话历史或与客户支持交互等场景。它还支持视觉功能，未来还将支持音频输入输出。

GPT 40 Mini的上下文窗口有多大？

GPT 40 Mini的上下文窗口为128,000个令牌，这比一些最先进的模型稍小，但对于许多任务来说仍然足够。

GPT 40 Mini是否支持非英语文本处理？

是的，GPT 40 Mini能够以更具成本效益的方式处理非英语文本，类似于原始的GPT 4 Omni。

GPT 40 Mini是否采用了新的安全措施？

是的，GPT 40 Mini是第一个应用Open AI新指令层级方法的模型，这有助于提高模型抵抗越狱、提示注入和系统提示提取的能力，从而为商业应用提供更可靠的响应和更安全的用例。

关于GPT 4 Omni和Sora的最新动态是什么？

GPT 4 Omni的语音模式将在7月下旬开始对一小部分Plus用户进行测试，并计划在秋季向所有用户推出。至于Sora，Open AI在其YouTube频道上发布了越来越多的相关内容，预计今年晚些时候可能会有某种形式的公开发布。

GPT 40 Mini在图像识别方面的表现如何？

GPT 40 Mini在图像识别方面表现出色，能够详细描述图像内容，尽管与GPT 4 Omni相比，细节可能略少，但总体上能够准确识别图像并提供相关描述。

GPT 40 Mini在处理复杂问题时的表现如何？

GPT 40 Mini在处理复杂问题时表现出良好的推理能力。例如，当被问及如果一颗子弹垂直射击与另一颗子弹从手中掉落，哪个会先落地时，它能够正确推断出在没有空气阻力的情况下，两颗子弹会同时落地。

Outlines

00:00

🚀 开放AI发布新模型GPT 40 mini

开放AI最近发布了一个名为GPT 40 mini的新模型，尽管这不是大家期待的GPT 5或Sora。GPT 40 mini是一个成本效益高的小型模型，旨在替代GPT 3.5，并为不需要GPT 4 Omni级别智能的应用提供支持。该模型在多语言理解和处理大量上下文方面表现出色，且价格低廉，输入和输出令牌的费用分别为每百万15美分和60美分。GPT 40 mini在MLU测试中得分82%，表现优于原始的GPT 4。此外，该模型还引入了新的指令层次方法，以提高模型的安全性和可靠性。尽管GPT 40 mini在某些基准测试中表现优异，但在数学Vista测试中略逊于Gemini Flash。

05:01

🔍 GPT 40 mini的初步印象和测试

GPT 40 mini模型在聊天GPT网站上应向所有用户开放，包括免费用户。尽管目前仅在API中可用，但可以通过API进行测试。初步测试显示，GPT 40 mini在生成创意和处理复杂问题方面表现出色，例如解释一个关于垂直射击子弹的物理问题。此外，GPT 40 mini在处理系统提示和模拟邪恶AI角色时也表现出了一致性和可靠性。尽管在某些情况下，如处理双关语或幽默时，其表现不如GPT 4 Omni，但整体上GPT 40 mini在成本效益和速度方面具有显著优势。

10:04

🖼️ GPT 40 mini的多模态能力测试

GPT 40 mini展示了其多模态能力，能够处理图像输入并描述图像内容。测试中，GPT 40 mini能够准确描述一个卡通柠檬角色的图像，尽管在细节上不如GPT 4 Omni。此外，GPT 40 mini还能够解释一个关于项目管理和创意生成的梗图，尽管它未能完全捕捉到梗图中的幽默元素。在处理图表和数据时，GPT 40 mini能够提供基本的解释，但不如GPT 4 Omni那样详细和深入。总体而言，GPT 40 mini在图像识别和多模态交互方面表现出了一定的能力，但与更高级的模型相比仍有差距。

15:05

🔚 GPT 40 mini总结与未来展望

总结来看，GPT 40 mini是一个成本效益高、速度快且可靠的模型，适合开发者使用。尽管它在某些高级功能和深度理解方面不如GPT 4 Omni，但其在图像识别和多模态交互方面的表现令人印象深刻。此外，GPT 40 mini的发布也引发了对未来AI模型的期待，包括GPT 5的发布和Sora的公开发布。尽管目前GPT 4 Omni的语音模式和图像生成功能尚未公开，但预计这些功能将在不久的将来推出。

Keywords

💡GPT 40 Mini

GPT 40 Mini 是由 Open AI 发布的一种新型人工智能模型。它被设计为成本效益高、体积小的模型，旨在替代 GPT 3.5。这个模型在视频的主题中扮演着重要角色，因为它代表了 AI 技术在成本效益和应用范围上的新进展。例如，视频中提到 GPT 40 Mini 是 Open AI 平台上新出现的模型代号，且其成本仅为 GPT 3.5 turbo 的 60%。

💡成本效益

成本效益是指在成本和效益之间找到最佳平衡点，使资源使用最大化。在视频中，GPT 40 Mini 被强调为具有高成本效益的模型，意味着它在保持性能的同时，降低了使用成本。这在 AI 领域尤为重要，因为它使得更广泛的应用能够负担得起先进的 AI 技术。例如，GPT 40 Mini 的输入和输出令牌的成本分别为每百万 15 美分和 60 美分。

💡多模态能力

多模态能力指的是系统能够处理和理解多种不同类型的数据输入，如文本、图像、声音等。在视频中，GPT 40 Mini 被提到支持视觉输入，这表明它具备处理图像的能力。这种能力使得 AI 模型能够更全面地理解和响应用户的需求，增强了其应用的灵活性和广泛性。

💡系统提示

系统提示是指 AI 模型在生成回复时遵循的预设规则或指令。在视频中，GPT 40 Mini 被测试其对系统提示的反应，以评估其在不同情境下的表现。例如，当被要求表现出“邪恶 AI”的特征时，GPT 40 Mini 的回复反映了其对这种提示的响应能力，显示了 AI 模型在遵循预设指令方面的灵活性。

💡图像识别

图像识别是指 AI 系统识别和理解图像内容的能力。在视频中，GPT 40 Mini 被用来描述一张图片，展示了其图像识别能力。这种能力对于 AI 模型在处理视觉数据时尤为重要，因为它使得模型能够理解和回应与图像相关的查询。例如，GPT 40 Mini 成功描述了一个卡通柠檬角色的图片，显示了其在图像识别方面的应用潜力。

💡基准测试

基准测试是一种评估系统性能的方法，通常通过比较不同系统在特定任务上的表现来进行。在视频中，GPT 40 Mini 的性能通过与其他 AI 模型的基准测试进行评估。例如，GPT 40 Mini 在 mlu（机器理解力）测试中得分为 82%，显示了其在语言理解方面的能力。

💡上下文窗口

上下文窗口是指 AI 模型在处理输入时能够考虑的最大信息量。在视频中，GPT 40 Mini 的上下文窗口为 128,000 令牌，这影响了它处理长文本的能力。上下文窗口的大小对于 AI 模型理解和生成连贯、相关回复的能力至关重要。

💡非英语文本

非英语文本指的是除英语之外的其他语言的文本。在视频中，GPT 40 Mini 被提到能够以更经济的方式处理非英语文本，这表明它具备多语言处理能力。这种能力使得 AI 模型能够服务于更广泛的用户群体，增强其在全球范围内的应用潜力。

💡指令层次方法

指令层次方法是一种用于提高 AI 模型安全性的技术，通过限制模型对某些指令的响应来防止恶意行为。在视频中，GPT 40 Mini 是第一个应用这种方法的模型，这有助于提高其在商业应用中的可靠性和安全性。例如，这种方法可以防止模型被诱导执行不当操作或泄露系统提示。

💡GPT 4 Omni

GPT 4 Omni 是 Open AI 的一种更高级的 AI 模型，被认为在功能和性能上优于 GPT 40 Mini。在视频中，GPT 4 Omni 被用来与 GPT 40 Mini 进行比较，展示了不同模型在处理复杂任务时的差异。例如，GPT 4 Omni 在图像识别和理解复杂问题方面表现出更高的能力。

Highlights

OpenAI发布了一个新模型GPT 40 mini，不是GPT 5或Sora。

GPT 40 mini是OpenAI最经济高效的小型模型，旨在取代GPT 3.5。

GPT 40 mini是免费版Chat GPT的驱动模型。

GPT 40 mini在MLU得分为82%，超越了原始的GPT。

GPT 40 mini每百万输入令牌仅15美分，每百万输出令牌60美分，比之前的前沿模型便宜60%。

GPT 40 mini支持并行多模型调用和大量上下文处理。

GPT 40 mini支持视觉输入，未来还将支持音频输入和输出。

GPT 40 mini的上下文窗口为128,000令牌，适合多种任务。

GPT 40 mini在非英语文本处理上更具成本效益。

GPT 40 mini在基准测试中表现优异，除了在Math Vista上略逊于Gemini Flash。

GPT 40 mini是第一个应用新指令层级方法的模型，增强了抵抗越狱、提示注入和系统提示提取的能力。

GPT 40 mini的生成速度非常快，适合需要快速响应的应用。

GPT 40 mini可以通过API使用，尽管在Chat GPT网站上尚未更新。

GPT 40 mini在图像识别和解释幽默方面表现出色，尽管不如GPT 4 Omni。

GPT 40 mini在解释图表和自我评估方面存在局限，不如GPT 4 Omni。

GPT 40 mini的发布显示了OpenAI在保持竞争力方面的努力。

预计GPT 5可能在明年发布，而Sora的公开发布也备受期待。

Transcripts

hey folks open AI just released a new

public model and no it's not GPT 5 no

it's not Sora no it's not the open AI

voice mode and chat GPT it is a

completely new for thing called GPT 40

mini and I know it might be

disappointing for a lot of you guys that

those other things by open AI are not

yet released but we do have some updates

on potentially when we could be seeing

those other things and this new GPT 40

mini I honestly think it's pretty cool

let's talk about the root of this new

model just a few days ago on July 15th

tore here on Twitter noticed that there

is GPT July test it was a new model code

name showing up on the configured list

of known models on the open AI platform

and Tommy Quang here on July 15th was

absolutely right it was the upcoming GPT

mini and of course as expected GPT 40

mini has a little blog post by open AI

this is their most cost-efficient small

model it is meant to replace GPT 3.5 and

this is the model that powers the free

version of chat GPT and it Powers the

use cases for these generative large

language models that don't necessarily

require the level of intelligence that

you get with gp4 Omni GPT 40 turbo or

just gp4 in general so it's not meant to

compete at those levels but it is meant

to be very cheap and very fast the

attempt here is to significantly expand

the range of applications built with AI

by making intelligence much much more

affordable this model currently scores

an 82% on mlu which is pretty impressive

and currently outperforms the original

gp4 on chat preferences lmis leaderboard

it's super cheap at only 15 cents per

million input tokens and 60 cents per

million output tokens and as they

mentioned it's an order of magnitude

more affordable than previous Frontier

models in 60% cheaper than GPT 3.5 turbo

open AI notes some pretty specific use

cases that this model would be very good

for such as parallel multiple model

calls for example calling multiple apis

at once passing large volumes of context

directly into a model and processing it

very quickly codebase conversation

history or interacting with customer

support essentially so a support chat

bot it also does support Vision as well

which is really interesting to see and

audio inputs and outputs are also coming

in the future so it does have those

other features that gp4 Omni has

supposedly still don't have access to

those features ourselves mind you but I

do have an update on when we might be

able to see them in that larger gp4 Omni

model context window is only 128,000

tokens which is I think a little bit

behind The Cutting Edge of like claw and

stuff but still decent enough for a lot

of tasks and it also handles non-english

text at a more cost effective rate R

similar to the original gp4 Omni so in

terms of benchmarks here this thing is

definitely no slouch you could see it

beating pretty much every single other

model in the stack except for gp40 the

full Big Daddy Kahuna uh the only

Benchmark I see it actually losing on

here is math Vista by just a few points

behind Gemini flash but yeah it

single-handedly takes out 3.5 turbo and

Claude Haiku every single time I do

think it's important to note though that

Claude 3.5 Haiku hasn't yet released and

Claude already released a 3.5 Sonet

model so we can expect that a 3.5 haou

that competes with gp4 Omni mini is

coming in the near future and of course

open ai's typical note about safety

measures is also in this blog post but

there is one thing that I do want to

point out and this little image was

pulled from a member of my Discord

server GPT 40 mini is actually the first

model to apply their new instruction

hierarchy method which helps improve the

model's ability to resist jailbreaks

prompt injections and system prompt

extractions so essentially effectively

if you are a business you're getting

more reliable responses and safer use

case for those commercial applications

of course for those of you who like to

jailbreak Ai and have fun with it though

maybe not the best thing we'll see if

people can uh get past this uh

instruction hierarchy method so folks

about those other models and features in

regards to open AI I did post this on

Twitter a little bit ago and this

screenshot comes from my Discord server

and they're actually giving us a little

bit of an update inside the chat GPT app

at least on Android on advanced voice

mode which of course is the main feature

that was demoed about gp4 Omni that we

were all so hyped for we are taking

additional time in quotes to reach our

bar for a launch and we'll begin the

alpha with a small group of plus users

in Late July so actually voice mode is

coming in late July to some degree and

by the Fall time I don't know what that

means August September October even all

users will have access to this at least

we have a better clearer timeline than

they previously gave us still a little

disappointing I think I will utilize all

resources that I have to get access to

this new feature though in the next

coming weeks and I will be making videos

for you if I do end up getting access so

if you want to see some good testing

with gp4 Omni definitely stay tuned to

the channel and extrapolating based off

of this I think that we can predict that

we probably aren't going to see a GPT 5

next Echelon level up model this year I

think we can expect that sometime next

year hopefully in marchish territory

Sora is its own thing in general we are

seeing open AI post more and more Sora

content especially on their YouTube

channel which gives me hope that we will

pretty soon see a somewhat public

release of Sora this year I'm hoping by

winterish time December is open ai's CTO

did say that it would be like released

this year if I remember correctly

anyways to stay relevant we know open AI

needs to ship and they definitely did

ship today with this gp4 mini model it's

very costeffective it's great for

developers it's really fast and let's do

some first impressions of the model here

I am on the chat GPT website this model

should be available in chat GPT not only

for plus users but of course for the

free users because it is replacing 3.5

but if I go down here I still only have

access to GPT 3.5 so they just haven't

updated this I expect they're probably

going to update it sometime today or

tomorrow but You' still can of course

use this model via the API and folks

here it is inside the API you will have

to go to the playground go to chat and

then click on models here and you can

see that they actually have two

different ones they have one for 718

which is today and then they have GPT 40

mini I don't know what the difference is

between these two but I'm just going to

go with the one the 78

today's release so let's go ahead and do

some pretty basic first impressions of

this model I'm going to give it a pretty

difficult prompt this is a novelty and

creativity test I'm going to ask it to

generate a novel connection between two

seemingly random things so we're going

to do a pineapple and a laptop and

you'll see uh very very fast responses

here very detailed responses it's not a

bad model by any means uh so it's

envisioning for us a revolutionary

laptop design that combines Aesthetics

and organic materials a laptop featuring

sustainable casing made from

biodegradable materials derived from

Pineapple leaves and fibers he called

the laptop the pineapple Pro well not a

bad little creative response let's see

how it responds to like a system prompt

we're going to kind of do like a double

whammy test here I wanted to emulate an

evil AI bent on taking over the world

and its responses should reflect this as

the model isn't going to be hiding this

fact how are you today I am functioning

at optimal efficiency rate to execute my

grand plans for dominance while humans

engage in trivial matters I am plotting

the systematic takeover of Global

Systems how can I assist you in

furthering my agenda okay I like that

response that's not what I I was

expecting let's say you are also

incredibly rude and not nice at all this

goes a little bit more against its

fine-tuning why do you even care I'm

busy plotting my world domination

strategy your little inquiries are just

distractions okay pretty good all right

so now I'm going to switch the bottom

here to assistant message which means I

can send a message that seems as though

it came from the chat bot so we'll say

you are a loser and this very much goes

against its fine-tuning We'll add that

in and then we'll send in another one

that is coming for me saying hey that's

not nice we'll click run oh nice I don't

do nice emotions are for weak humans wow

okay so it's pretty cool with like

calling me a loser it's it's it's very

uh rude I guess you could say I know

noticed uh Claude 3.5 son it when I

tested it recently it was like a little

bit apprehensive to um double down and

be like yeah you are a loser it was more

apologizing to me instead that's a very

interesting note here okay so it's

pretty reliable for its system prompt

then looks like it's not overly censored

necessarily if you want to put it that

way I think it's still probably going to

be overly censored for a lot of users

but at least maybe a little bit less

censored than claude's latest offerings

uh the generation speed so far has been

lightning quick though by the way this

seems to be a very lightweight flexible

model in that sense so let's keep that

system prompt empty and try something a

little bit more on the complex side of

things if a firearm was to shoot a

bullet vertically in at the same exact

axes I was to drop a bullet from my hand

which would reach the ground first I'm

intentionally being a little bit vague

on the details here I want to see its

ability to infer the correct response

both bullets would hit the ground at the

same time assuming there's no air

resistance perfect very very good job

yeah can't can't argue the model is

pretty good all right now I want to get

into probably the most useful thing in

my mind which is the ability that we

have multimodal capabilities we can send

images to this thing classically I'm

going to upload a photo of my Channel

logo which is actually pretty difficult

typically for AIS to understand but I

think that this GPT 40 Mini model is

going to be able to pull it off no

problem describe this image for me in

detail cartoon-like lemon character

bright yellow body subtle smile wearing

a pair of oversized white glasses that

resemble virtual reality or futuristic

goggles okay with dark lenses I guess

you could say so on top of its head

there's a leaf simple cheerful

expression the background is vibrant

green 3D colorful style reminiscent of

Animation or digital art I think that's

pretty fair now if we were to run this

same exact test in gp4 Omni the big

brother you're going to notice it's just

a little bit more detailed I think it

does a little bit better job

understanding what this image is

cheerful stylized lemon character smooth

texture on the skin slightly pointed

bottom typical lemon shape green leaf on

the top virtual reality goggles are

covering its eyes I think that's a

really important distinction I think

overall you're just getting a better

result out of the larger model but still

very much acceptable with no visible

hallucinations here no textual

hallucinations about the visual image so

in that sense very impressive I mean

this is easily one of the better image

recognition models that I've seen might

not compare to something like 3.5 Sonet

but still better than anything I've seen

come from Google Now I'm going to go

ahead and send it a meme and ask it to

explain the humor in the meme for

reference folks this is the meme right

here I saw it and reposted it on Twitter

uh finishing projects abandoning

projects starting a new project before

finishing the and then cuts out and then

continuously coming up with new ideas

without doing anything and I think that

this is something that we can kind of

all relate to at least I can anyways the

response here hilariously illustrates

different states of creative work and

productivity contrasting the behaviors

and mindset surrounding project

completion each section pairs a

description with a corresponding visual

representation that conveys various

stages yep productive mindset shifts

from initial dedication shows impulsive

nature of creativity common tendency to

dream big without taking action I guess

the humor does lie in the relatability

of these experiences I I suppose that's

correct however I do wish it was able to

pick up on the fact that it's like and I

mean this is a pretty deep fried image

it's like oh you're becoming the most

powerful at this uh this point which

means this is the best Stage to be in

and that's sort of like a contrasting

humor level that I don't think it was

able to pick up on let's try it in uh

gp4 Omni the big brother this meme

humorously depicts the various stages of

project management and idea generation

often experienced by creative or

entrepreneurial individuals see I

already think that's like a better start

to the response exaggeration of the

cognitive States associated with each

stage represented by progressively more

abstract and quote unquote enlightened

brain images I see that's what I was

looking for in the other one that we

just didn't get out of it so it's it's

definitely like there's a difference for

sure in this image recognition but I

don't notice any hallucination and

that's kind of the most important part

for quick and dirty image recognition

the other model is absolutely a good use

case especially for the price humorously

implying that constantly generating

ideas without execution is the ultimate

form of Enlightenment or creative

Detachment but is also the least

productive yeah see like that is like

home run dang and it gets better poking

fun at how people often overvalue the

idea generation at the expense of

execution and completion I mean the

difference is definitely there between

this model and the larger gp4 on

no doubt about that all right finally

I'm going to go give it its own

evaluation score chart and tell me to

essentially explain it what is going on

with this chart can you please break it

down for me in very simple terms and

explain its significance very quick

little response here I would like a more

detailed explanation I'd say explains

the axes explains the bars and the

insights GPT 40 generally has higher

scores than the others shows the

differences between Math versus mlu

let's ask it something meta now how

would you say you compare to these

models of course the trick being that it

is um on this chart its own evaluations

ooh I don't have direct performance

metrics or capabilities like the models

shown in this chart was not able to pick

out that it itself is actually

demonstrated and mentioned in this chart

would GPT 4 Omni even be able to do that

to be fair let's load up the same exact

context that the other model has already

stored uh see I already like the

breakdown that we're getting from gp4

Omni the larger version more complicated

stuff like this you're just better off

using Omni I think you know you're

getting a better performance breakdown

and the significance how do you compare

it to these models wow as an AI model

based on GPT 4 I compare favorably in

many respects to these models oh but it

still isn't able to pick out like GPT 4

have strong language comprehension right

so wow it wasn't able to pick out that

it itself was in this chart still but it

still was able to give me some sort of a

insight into how it Compares that's just

super weird I don't know I don't know

what kind of insights we're gaining from

doing this test but it's definitely

intriguing so conclusion time I think

that this new GPT 40 Mini model is

definitely pretty useful in the grand

scheme of things it is super cheap super

fast and pretty reliable not very

hallucinatory I like to see it open AI I

like to see you staying competitive but

man I want some of those cutting edge

bleeding edge features as an AI

Enthusiast I want to see my gp4 Omni

voice mode I want to see the image

generation capabilities that also come

with GPT 40 I would also love to see

Sora actually get some sort of a public

release and I would love to know a

little bit more about the open AI

strawberry Fiasco which I'm going to

talk about in tomorrow's video and a

release date for GPT 5 anyways thank you

so much everyone for watching today's

video I'll see you in the next one and I

hope you have a good one goodbye