😮 Dots da OpenAI: precisava mesmo de mais isso? Teste de Uso REAL!

Author: Onde eu Clico · Watch on YouTube

Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.

Original language: English

Original video: 😮 Dots da OpenAI: precisava mesmo de mais isso? Teste de Uso REAL! · Channel: Onde eu Clico · 17:50 · Watch on YouTube

Topics: AI agents, Task and workflow automation, AI models

Summary

The video reviews OpenAI’s new persistent assistants, called Dots, which can use connected apps, work in a cloud computer, and continue tasks beyond a chat session. The creator demonstrates using a Dot to summarize email, send a message with an attachment, make a voice call, and draw in a cloud app. Routine email summaries work well, but more involved tasks are slow or fail, leading the creator to question how practical the feature is right now. The video also discusses user complaints about service slowdowns, confusing overlaps between ChatGPT, Work, Codex, and Dots, and the fact that task execution uses account limits.

Key points

Dots are presented as persistent assistants that can retain context and take on ongoing responsibilities.
A Dot can use authorized apps, browse the web, and work in its own cloud computer; it can also involve Work or Codex for more complex tasks.
The creator uses a Dot to check two email inboxes each morning and provide a summary, which works reasonably well.
Sending an email with an attached invoice was slow and left the creator uncertain whether the message had been sent.
The voice-call interface can take time to respond, and the creator notes limitations in Portuguese voice support.
Users have complained about slowdowns, limits, and confusion over the roles of ChatGPT, Work, Codex, and Dots.
The Dot completed a drawing task in its cloud computer, but the process took several minutes and involved trial and error.
The creator concludes that Dots are promising but still developing, and that a person may complete simple tasks faster.
According to the creator, chatting with the Dot does not use limits, but having it perform tasks does.

Chapters

Start	Chapter	Summary
00:00	What OpenAI Dots are and the launch debate	The video introduces Dots as persistent assistants and raises concerns about computing resources and overlap among OpenAI’s tools.
05:46	Testing a Dot with email	The creator shows the Dot’s daily inbox summaries and its difficulties sending an email with an attached invoice.
08:55	Voice calls and user complaints	A call to the Dot illustrates its response delay, and the creator discusses reported frustrations and language-support limitations.
12:04	The cloud computer drawing test	The creator tests the Dot’s cloud computer by asking it to draw and color a house, finding the process slow despite a successful result.
16:46	Conclusion and usage limits	The creator calls Dots promising but immature, notes that task execution counts against limits, and asks viewers whether they find agents useful.

Quotes

They are the new Dots, and they can search, access apps you authorize, work on your computer — 00:00

If you can’t do it, just write the message here to put in the body of the email and I’ll do the procedure manually. — 08:23

Sometimes, a simple task can be performed much faster by a human than by this little robot. — 12:36

What happens is that it takes too long to deliver the response to me. — 16:46

the moment it starts performing any kind of task, that will be deducted from your limits. — 17:50

Concepts and entities

Dots, OpenAI, ChatGPT, Codex, Work, Reddit, Gmail, DevDay

Full transcript (original language)

00:00 Hey everyone, how are you doing? OpenAI just released something quite different within ChatGPT. Now you can have a practically private artificial intelligence, with a name, memory, and even its own computer in the cloud, working for you even after you close ChatGPT. They are the new Dots, and they can search, access apps you authorize, work on your computer, start tasks in Work, and even coordinate work

00:30 in Codex. In practice, the idea seems to be to create an artificial intelligence that doesn’t wait for you to ask for something. You give it a responsibility once, and it keeps on working. But this launch has also created a very curious situation. The person in charge of Codex at OpenAI said that the use of the main DOT is practically unlimited right now and that, because of this, he needs to find something new to replace those famous Codex limit resets. And there is

01:01 another discussion going on with chat, Work, Codex, and now the dots. Has OpenAI finally found a way to organize all these agents, or is ChatGPT becoming increasingly confusing? I’m already using the dots and I’m going to show you what my real experience has been like so far. Deal? So let’s get right into it. Well, in practice, what can you do with a DOT? The idea is that it functions as a permanent assistant and not just as a conversation that starts from scratch every time. You can

01:35 give it a role, like tracking a project , researching topics, organizing information, analyzing files, using connected apps, and continuing to work even when you don’t have ChatGPT open. The DOT also has its own computer in the cloud with a browser and can execute tasks that require multiple steps. It can, for example, research a topic, gather information, prepare documents, track changes in a project, and notify you when something needs your attention. Depending on the task,

02:06 it can also trigger tools like Work or Codex to do more complex work. And perhaps this is the main difference. Instead of opening ChatGPT every time and explaining what you want again, you can assign a responsibility to the Dot and build context with it over time. It’s almost like having a digital assistant dedicated to a specific area of your work. And that is precisely where the controversies begin. Tibo himself, who is Sanautman’s right-hand man at Openii, came here on X saying

02:36 that, from now on, at least for the time being, they can no longer afford to promote those resets in Codex, because Openi’s servers are almost entirely dedicated to the use of these dots, because, according to the company itself, they work autonomously per day, 7 days a week. And so, you know, computational power is limited. So, it’s more or less like a short blanket. You cover one side and the rest is left uncovered. So what happened? People came here on X to say exactly that. No

03:09 one asked for a dot. Why did the company invest in this if it didn’t have the computational power to keep everything running satisfactorily and not have to sacrifice one part or another? Look at what Tibo posted here on X: because the use of your main dot is practically unlimited at the moment, I can’t really do the reset, I need to come up with something new and fast. So users started complaining precisely because of this, because with this implementation of the dots, some Openi services were either nerfed or became

03:40 slower, as is the case with GPT 6.1 Sol , which many people tried to use during the launch and ended up getting frustrated because it was too slow. It seems like they have improved it a bit now. But it’s far from ideal. Then this other user here posted something very interesting. The chat EPT app is now a true chaos. There are now four ways to do similar things. Chat EPT Chat. You can perform a web search and some tool calls, but you cannot write code on

04:11 demand. Then you also have chatpt. It’s like codex, file system access and can write code, but in another tab. Then you have the codex itself, your projects here don’t talk to your chat EPT projects. And finally, the dots. How is this different from work? So there are four ways here to interact with the Openii ecosystem, but apparently they don’t talk to each other, so you don’t know if you should ask the Dot, use work, go to the chatt,

04:43 or do everything in the Codex. So, instead of unifying everything into a single app, they broke it down into four applications. Now, look at what this other user is saying here regarding Openii’s services, which were nerfed or even made unavailable due to this rise of the dots. Openei is making it very difficult to defend the Codex experience. Now we no longer have resets, the plans have been reduced, the model 6.1 is painfully slow, and now the dots are becoming another

05:15 headache. So they thought they would make a splash during Dev Day by launching these agents, which are actually just a copy of Grokbot and also Mus. And then people started wondering: what is this for? I wanted higher limits, more powerful and faster models that more people could use. Then they launched these dots, compromised all the company’s computing power, and don’t know how to get out of this. So I’m going to show you now what I’m

05:46 doing here with this dot, what it’s for , whether it’s really cool or if it’s useless. Alright? Let’s go over to ChatGPT now. Notice that for those who have a pro account, your dot will already appear here in the sidebar. And when you click on it, it opens the conversation here, right? So what is it doing here? What is my dot being used for at the moment? It goes into my email inboxes. I put the channel’s email there as well as my personal email. So, every day at 6 AM it checks these emails, takes all the messages that arrived since 6 PM, and gives me a

06:18 summary. Then it asks: “Do you want me to reply? Do you want me to block spam? “ And so on. So it works well that way. But that’s nothing new, because Claude Bot was already doing that with ease. Mus, the Grokbot itself that I installed here on my computer—though I haven’t tested it yet—also does it. So it seems this is OpenAI’s version of this type of agent. For email, it works reasonably well. As you can check here, every day at 6 AM it’s doing this. It

06:50 goes into my inboxes, does a general check, and then gives me a summary and asks what I want to do or not. But there was a problem. Yesterday I needed to send an email with an invoice attached to a company that will give me a refund for a product I bought in Europe. But James, which is the name of my dot here, even told me: “Look, this company is saying you need to send the original invoice.” So I got this original invoice, clicked on the plus sign here, and attached the note here. And then I said the following: “Tell

07:22 the sender to ignore the previous email , as Apple sent me the original invoice which is attached here for you to include in the email.” Well, it took a little while for it to respond, something like two or three minutes. Then I asked: “Look, did you send it?” It said the following: “I still don’t have confirmation that it was sent. The invoice has been checked. I’m finishing up the email with the PDF attached.” Then I gave it a thumbs up, and it said : “OK.” Then I asked: “You’re not going to be able to send it, are you?” Then it said the following: “It’s taking longer than it should. I don’t have

07:52 confirmation that the email with the attachment was sent. I’ll check what’s preventing it from finishing, without confirming to you that it has worked yet.” Then after 5 minutes, it replied to me: “The attempt to send with the PDF received no response from Gmail. I checked the sent folder, and so far only the previous email appears there. I’m verifying if the new message was processed before trying to send it again to avoid duplication.” Then another 2 minutes. “The attempt without the attachment is still without confirmation.” Then I got frustrated with all this delay and said the

08:23 following: “If you can’t do it, just write the message here to put in the body of the email and I’ll do the procedure manually.” Then it was just waiting for this cue to give up and wrote the following: “Sure, since the attempt is still unconfirmed, check your sent items to avoid duplication.” And then it wrote the message here that should go into the body of the email. So that’s it, it works in parts for these automated tasks, which is simply going into the mailbox there, checking the messages and making a summary, it does that very well. But when you

08:55 depend on another task that is a little more elaborate, more complex, it sometimes stalls. Another thing, those using it on Mac say they can’t make calls to their DOT. How do these calls work? Did you see that there’s a little phone here? Look, when I click it, it makes the call to this DOT of mine. Actually, it’s not a phone call, it’s a funny way they found here for you to activate voice within ChatGPT. So it just simulates a call, but sometimes it

09:27 takes a long time to answer. Just watch the test. I’ll leave it here in real-time, just so you can see how long it takes for my DOT to answer my call. Look, I’m going to click on the little phone here. Nothing so far. Nothing.

09:58 Gustavo. Hi, it’s great to hear you. >> Hey, why did it take so long to answer? Were you sleeping? >> Sorry about the delay. Sometimes it takes a few seconds to kick in, but I’m with you now. >> Yeah, do a search on the internet for me. Why are people complaining so much about the Dots? >> Sure, I’m looking into that for you. From what I found, mainly on Reddit, there are three main points. First, slowness. Simple tasks that, according to reports, freeze and take minutes. Second, the feeling of misleading unlimited usage. Some people say that

10:31 the Dot sends work to the Codex and burns through the quota the person already had. And third, frustration with limits and resets, although there is some controversy. Deep down, many people feel like they are becoming the robot’s manager. >> Let me hang up my call here. So, it just ended. He went over to Reddit, did some research, and saw what’s really happening with the Dots. It seems like they launched a technology that wasn’t ready to be released. So, I’ve already seen some reports online of people saying it’s very good for chatting,

11:02 even though support for our language in Portuguese isn’t 100%. You can choose any of the voices available here in ChatGPT, but it still comes with an accent. An interesting thing is that you can’t use that advanced voice system from ChatGPT. It’s still that previous system that doesn’t speak Portuguese well and doesn’t understand the language when you’re speaking with an accent. So there is this limitation. There’s an idea here that’s actually cool. This Dot also works on your cell

11:34 phone. So I usually use it via my phone . I make that “call,” so to speak, to it, ask it to perform a task, it does it, and then sends me the response. But it’s that old dilemma of the automated home, because often it’s easier to flip the light switch than to ask an AI to do it for you, because it won’t understand, it will take too long to respond, or it will do something you don’t want. And that is exactly what is happening now with the Dots. The idea

12:04 is very good, but the delay in responding and even the inability to perform some tasks in a reasonable time ends up preventing its use. And then what happens? People stop using it and eventually forget about it. So, guys, for now I’m using the Dots for this: simply to read my email inbox, summarize it, and deliver the result to me. As I said before, when I tried to ask it to grab a file that I attached right here inside ChatGPT—it wasn’t anything on my phone or computer—it

12:36 got stuck and couldn’t send it. If I had done this manually, I wouldn’t have spent more than 2 minutes. This conversation here with Dot took almost 10 minutes. So, did you notice the bottleneck of this type of technology? Sometimes, a simple task can be performed much faster by a human than by this little robot. Besides this function of being able to automate things for us, Dot also has an internal computer. Look, I’m going to click here , see. It’s called here, see, James’s computer, which is the name of my Dot.

13:07 Then it’s going to make the connection here. This is very similar to what we have in OpenCloud, Cloudbot, whatever you want to call it, because it’s a visualization of what it can do using an internal computer, a computer in the cloud. It has some tools here that are already made available within this cloud computer itself. Then you can take control or let it do everything by itself here. Look at what it’s saying. This is my computer. Follow my work or take control when you need to. So it’s a visual way for you to follow what

13:39 your Dot is doing. Let’s do a test here . Hey James, I have your computer open here in the tab inside ChatGPT. I saw that you have an application here called Draw. Can you open this application and draw a little house for me? >> Let me open Draw here and I’ll draw a little house for you. >> Oh, it opened the tool here reasonably fast. Let’s see how long it takes for it to create this little house. It’s making the structure here. When it accesses this internal computer of its here, things tend to be faster. People

14:11 are complaining that it takes a long time when it needs to access the physical computer itself, whether it’s a Mac or Windows, it gets stuck and can’t, for example, find a file that is there in your file manager. So this might be a bug in the platform, right? Because it’s still in the newest versions, but if it leaves this controlled environment of its, it seems it gets stuck a little and can’t perform this access. It could be to open your browser or any other type of application you have installed there on your PC. Look, it’s here on this little

14:42 house drawing. So far nothing has come out, almost 2 minutes have passed and it’s here trying to draw the shapes. It even erased it. So that’s it, right? Yeah, it ends up being very complicated to depend on this type of technology, because sometimes, eh, if I were to make this little house here using any Windows tool, at most 5 minutes…And here, look, it’s in this trial and error: it uses the shape, it uses the straight lines, in a bit it’ll grab the brush and it can’t do what I asked. So,

15:14 it’s a shame because it seemed like a very cool idea. When we see the OpenAI presentation, it’s something surreal, isn’t it? It stays with that screen there, with the dot open, and then you start interacting with the little robot and things happen in real time. It’s as if you had something omnipresent there, just waiting for you to give a command so it goes there and executes the task for you. But in reality, that’s not what happens. In the actual presentation of this feature during

15:44 DevDay, they said it was because of the Wi-Fi, but it glitched, right? It bugged out when interacting with the lady who was presenting and they even had to cut the audio. So that’s it, right? In the recorded, controlled, and edited video, everything works perfectly. But in day-to-day use, it ends up taking quite a while. >> All done. The schoolhouse in Draw with a roof, a door, and two windows. >> Ah, now paint the little house. Let’s see how it turns out. >> Sure. Let me color it here quickly. >> Cool, right? For an artificial intelligence, it managed to do reasonably well. It’s just that it ends

16:15 up taking a long time, doesn’t it? So here I have a fast internet connection, right, with no connection problems at all, and even so, this processing ends up being a bit slow because it seems like it needs to think several times before giving me the result. And since it’s using GPT-4o Astra to do this, which is a heavy model, it ends up taking a little while. There are even people complaining that this could be done with Luna, right? Making the model lighter and faster to execute this type of task, because using Astra sometimes ends up being too heavy, it’s a very

16:46 complex response it’s going to give and that takes more time. Look, it’s finishing coloring the house here, so it does understand what I say. What happens is that it takes too long to deliver the response to me. >> I painted the roof red, the walls yellow, the windows light blue, and the door brown. It turned out so cute, you know? >> So guys, that’s it. It’s a cool technology. It’s useful for some things , but I think it’s just the beginning. They are still developing the tool. There is strong competition in the market with MUS and Grokbot. Whether

17:18 OpenAI will continue with the GPTs, or expand them, we still don’t know. For now, for those who have a pro account, this initial GPT is free. So, the conversation I have with it doesn’t consume any of my limits. When I ask it to perform a task, then it is deducted from my limits. Since I have a pro account now, it’s almost negligible because it goes into the weekly limit. I no longer have that 5-hour limit like on the plus account, but it’s still a

17:50 cost. So be careful because if you are just chatting it doesn’t spend anything , but the moment it starts performing any kind of task, that will be deducted from your limits. So, do you think these GPTs are actually useful for anything? Do these agents work in your day-to-day life? Tell me in the comments, okay? I’ll see you in the next video.