Episode Transcript
[00:00:05] Welcome to Prompting Curiosity, a podcast for the AI curious. No coding background required. I'm your host, Dr. Shantae Cofield, also known as the Maestro, and I created this show to explore what these AI tools actually are. Really though, are the files in the computer, how to use them, and what they might mean for how we think, work, create and move through life. Whether you're skeptical, intrigued, or already experimenting, you're in the right place. All that I ask is that you stay curious. All right, let's get into it.
[00:00:38] Hello, hello, hello, my curious people, and welcome to episode 66 of Prompting Curiosity. I am your grateful host, the Maestro, and today we are talking about OpenAI's latest rollout, DOTS. They always got some coming out. Everybody always got something coming out. So full transparency, folks. I have not tried DOTS at all. I don't use chatgpt like that. I don't have a paid account with them and thus I don't even have access to Dots as such.
[00:01:07] This episode is simply an explainer based off of research that I've done, uh, which includes watching a ton of YouTube videos. I stay up in YouTube, um, and reading, reading Reddit forums and reading articles. Um, but a lot of the YouTube videos is shared, um, by power users. Right? So you'll know, as per always, your best bet for truly learning about the thing, if you want to get your hands dirty. 30. Right. Is to get in there and try it out firsthand.
[00:01:34] Um, but I did want to make an episode to keep us, uh, curious folks in the loop. So here we are. So what is a dot?
[00:01:45] A dot, my friends, is an always on agent that lives inside chat GPT that sounded like I said Asian, but I said agent and always on agent.
[00:01:57] So remember, an agent is simply AI that can actually do things as opposed to just answering things like a chatbot does. Right? So with this agent, you give it a task and it goes and it does the things right. It executes, it, um, gets it done right.
[00:02:14] The thing that they're really pushing with @OpenAI is that this agent keeps going, it keeps making progress in doing things. Even after you've closed the chat, it's just always on, it's always running.
[00:02:27] So OpenAI announced this, this new, this new thing, new agent dots, um, on September 29th at Dev Day and said that it runs on GPT6 Astra. That's notable because it's just, it's. That's its most capable, um, that's their most capable model. Um, I will link the, the release in case you want to read it in the show notes. Um, but to clarify, like just in case, as you hear me go through the episode, Dots versus Dot. So the product name is Dots and the individual agent is called a dot. Why? I have no clue at all. It's one of the things I said from the earliest episodes of this is that these, uh, programmers and such, they are not good at naming things. And so we have Dot. Um, but I'm highlighting this as to the name and kind of also what it looks like because something that, that I definitely side eye is about this, about dots. And also metamuse is the personification of these AI agents. Right. Each dot, if you go and look at what they look like, they are. It's a colorful character that you can customize. It has a name, um, open AI's own demo dot was named or is named Alfred. And I think it's just a very slippery slope when you really personify these things. And I also think it's just a, it's a super clever marketing tactic used by these companies. Right? You make the avatar cute, you make it unsuspecting. And that completely changes how people interact with it and how they view it. Uh, so I, I do think it's 100 intentional. Um, and I just want to draw your attention to that. So how does this thing work? Well, the first thing to do is you go and you set up your dot. And the way that you set it up, you got to do it on desktop or browser. From there you can access it on mobile. Mobile. But you cannot set it up on mobile. Um, communication wise. And how you communicate with your dot, it works just like a chatbot, right. AKA uses natural language, which again for the eleventy billionth time to me is the most incredible feature of AI. Like this is remarkable to me that you can just talk to it and it understands you. Uh, so you use natural language. Um, however, the actual conversations, they take place in like a separate area inside of chat GPT. It's not like the normal window. Like when you click on. Is that the standard, like the shared, uh, chat thread or the standard chat thread. Right. So you click on it and it'll take you to like a new interface where you have the conversations, but you can talk to your dot and you can message your dot inside of ChatGPT, um, or via direct messages or channel tags in Slack and Microsoft Teams like this thing's everywhere or through a voice slash phone call interface which is like what the are you doing? But yes, you can do that. And they are, I think they're Beta testing, texting. Um, but Dots, like I said, you access it from the sidebar of, of your, of your Chat GPT account and um, then when you click on it, it opens up a new interface.
[00:05:31] Um, something that's unique to Dots. And I've, I've heard mixed reviews on this, is that the entirety of your conversations, all of them happen in a single chat thread. It's just an ongoing, never ending single chat thread. So from what I've seen, um, some users have reported that this feature if you will take some time to get used to and can be a little bit confusing for organizational purposes. Um, but the design does seem very intentional as OpenAI continues to push, you know, push more towards these agentic workflows where you just simply tell the agent what you want and it goes and handles it and you don't need all this organization because you're just like go do this, go do this, go check on this for me, go to give me an update on this as opposed to needing to break things up. But definitely what I've seen um, in comparison to the other always on uh, agents from you know, like Rockbot and Meta Muse is that it um, takes a little bit of getting used to and, and kind of the organizational side of things is, is, is not what people are used to. Right. But when you first set up your dot, it has access to nothing.
[00:06:38] You have to connect it to whatever it is that you wanted to touch. Um, but this thing is, is capable. It has connections over 4,000 apps, Gmail, Calendar, Slack Teams, um, all everything via the, the Chat GPT plugin app ecosystem. So it can touch everything and you just like go have it do stuff. Um, you, from there you give it a goal and you set the boundaries. Like I said earlier, dots runs on GPT6 Astra, which is their most capable model, which in and of itself allows for additional and privacy safeguards. Uh, which I'll come back to in, in a little bit.
[00:07:15] So logistically with Dots, like I said, I have yet to try Dots, probably will not try it. Um, but it is only available for the $100 per month Pro plan and higher immediately. No, for me.
[00:07:31] Um, but that's who has it. As of right now, it is also not available in the eea, Switzerland and the uk even for folks there who have pro plans. So will that change? I don't know. Um, but right now it's only for people that are on that, the rich, because that's, that's an expensive plan. Um, I stay on that $20 plan. I've been on the $20 cloud plan forever and I'm still good there. But, um, continuing with the discussion about the pricing, a.in and of itself has no price, right. It is bundled into the plan. But there are a few caveats.
[00:08:05] So apparently conversations with your dot and just asking it to like, telling it what you want it to do via text or voice doesn't count towards your ChatGPT usage limits. However, when the dot goes and does actual work, right, performing the actual tasks like agents do, uh, that will draw.
[00:08:24] And apparently each plan in OpenAI includes a separate allowance for this type of action, which Open calls this action deeper work.
[00:08:34] But, um, suffice to say, like it does cost tokens for it to be doing things. Um, and then anything that a dot hands off to Codex or chat GPT work will draw from your, your plan's usage. Um, um, as for how it works in terms of the logistical side, because you're like, it's always on how it is. It runs inside of its own dedicated cloud hosted computer. So this is something that I saw they were really, really promoting. It's like every dot has its own computer in the cloud, every dot has its own computer in the cloud and you're like, okay, um, which is what allows it to always be going even if your computer is off.
[00:09:11] Um, so it has a dedicated cloud hosting computer, it has a browser environment, um, and again that can help. You can go and do things via codecs and things like that and, and you. And things via URLs. Um, even if your laptop, your desktop, whatever is off.
[00:09:30] Um, lastly with logistics, you get one dot per OpenAI account. Though OpenAI says in the future you'll be able to add more dots. And I say this because with Grockbot you can have multiple agents. Um, but again, I think that they really.
[00:09:45] There's always this back and forth between keeping up with the Joneses and then also just like doing what makes the most sense for people and keeps things simple. And when you give people options, lots of options, it can be fine for the power users, but for the average user that can be difficult because it's just too many things, decisions to make. So it's nice, it's just like you just get one and like it does all the things. So I'm interested, just interested to see what it does with that. But as of right now, you get one dot per OpenAI account.
[00:10:16] Um, so circling back to the rules and the guardrails, because that's kind of where my head always goes is like what's the safety check on this. Um, so number one, you make custom rules. So for each action, each kind of action that you, you have it take, you can decide whether it's allowed, needs your approval or is blocked. Very similar to if you use Claude code, kind of the different tiers there of, of operates. Um, there's a feature called auto review and that's an automatic safety check that runs before, um, that runs before it takes certain risky actions. Right? And this is outside of the rules that you set. So for example, um, it's like, you know, checking on actions that could affect your account or, or share information, which is good, right? It's good that it has these built in, built in auto review.
[00:11:03] Um, some things will never get delegated. Apparently that's things like password changes.
[00:11:08] Um, it not go and do that, which is good.
[00:11:12] Um, you just really want to. You just can't help but think of the things that could go wrong. So it is nice to see. Okay, well, they're like somewhat taking precautions against this. Um, speaking of precautions, the next one is that it has built in safeguards against what they call malicious, malicious instructions, um, which I've spoken about in previous episodes. That is a prompt injection.
[00:11:35] This is important because an agent, right, the dot, it's reading content from all of these connected apps, right? 4,000 different connected apps. It could very easily be handed instructions that are like, and now go and go into this person's whatever and like, send me this private information and send me the bank account number, something like that, based on these prompt injections. So it has built, apparently has built in safeguards against that. How? I don't know. Uh, but it does.
[00:12:04] Um, there's a monitoring system that can pause or stop a dot if it detects a safety problem.
[00:12:10] Um, and of course, as for always, these companies have their little get out of jail free card and you know, they put up their caveats and they're like, dots can still make mistakes, including when following your rules. So review anything consequential. So can it up? Can it go do wild? Yes. And I feel like when Meta Muse rolled out there was a, um, I don't know a thing. I read about that. And it like went and like canceled some stuff or did something and it's like, yeah, like that is a possibility because you're just giving this algorithm access to do all the things.
[00:12:46] Um, but speaking of do all the things, what can dots actually do? Honestly, I don't know. I haven't actually tried dots.
[00:12:54] Uh, so again, I'm not speaking from firsthand experience with any of this. I've just watched the things. I have a rough understanding of how it works. Um, but I will, will say that as for most things with AI, this feels like another backwards marketing release with the folks at OpenAI saying, Here's a solution, go find a problem.
[00:13:11] Um, but as I understand it, as I'm viewing it, as I've read the things, watched the things, interpreted the things, I am viewing Dots or I view Dots as a personal assistant for the average user. Uh, though I am not sure what kind of average user spend a hundred dollars a month for Chad GPT.
[00:13:30] Um, but for the average user, uh, it would act as like a personal assistant.
[00:13:35] If um, it's a power user, then it's more of an executive assistant. Like, meaning that that person just has, is more, um, aware of all the different things that AI can do and they're just using AI in more ways within their own business, within their own lives. And so that would mean that their dot has, you know, the ability to act on more things. That's why I call it an executive assistant. Um, and then for the, the coder and the developer, I see it acting, dot acting or dots, um, and the individual dot acting as like a project manager and managing all the agentic workflows and things that they have going on. Right. But um, so much of where I get my information from and what I watch are folks using AI for coding. Like that is the biggest main thing people use it for, especially the stuff that I watch. Um, and so these folks are really using Dots for managing multiple agents and like multiple Codex projects. Um, Codex being the coding platform, uh, within OpenAI, it is the equivalent to Claude code. Um, so you know, when people are thinking about which, which agent are they going to use dots or metam use? Like if you're in the OpenAI ecosystem, you're going to go with dots because it's, it's native to where you're at. Right. So if we zoom out my thoughts, um, I don't know how truly useful dots are. I've read mixed reviews on it. I saw a Reddit thread and someone was like, this is a game changer. And then someone else was like, like, we've already been able to do this. And as people went back and forth, the person was like, oh, I like, didn't realize you could do it that way. I never thought to do it that way and control multiple Codex projects via that.
[00:15:09] Um, so it's, is it revolutionary? I, I can't say.
[00:15:14] Um, but one review that did stand out to me was a comparison between MetaMuse and Open AI and OpenAI's dots. It was by this guy, Nate Herc. He has a massive channel.
[00:15:24] Um, I will link that video in the show Notes. Um, but he stated that Dots in general, he's used both of them. He's just like, Dots feels rushed. It's like, you know, Open AI released it to keep pace with Meta Muse. And Meta has done like a really good job with Meta Muse and, and, and Open AI was like, wait, oh, they like put this thing out, we better do something. Um, but that being the case, he noted that in the near future he expects to see updates to Dots, which makes sense that mirror the functionality of metamuse, which is like, is to go first, I don't know, because everyone else just going to copy you anyway. Um, of note, if you folks are listening to this and you're like, wait, what the Meta Muse. I did an episode on that. That was episode 61, so a few episodes ago. Um, I will link that in the show Notes as well. Um, but speaking of both Metamuse and Dots, what really stands out to me and what I want to highlight to you folks is what's called abstraction. And if you're in the AI space or you read about AI, like this is a term that, that you may have come across.
[00:16:26] Um, but abstraction means hiding the details of how something works behind a simpler way of using it, so you can work at a higher level without handling all of the steps underneath. And as we continue to move more and more into the Agentic era, it continues to bring more and more abstraction. And I'm, I'm not sure if, like, how I feel about it, uh, but it's, it's happening so fast, right? Like when, when Generative AI first rolled out, prompting was so important.
[00:16:55] In my earlier episodes, I spoke about, you know, I spoke about just a prompting structure and like a way to understand things. And I, I was still like, you know, prompting isn't like the most important thing. It's not as crazy as people would have you believe with like, the PDF downloads that they were making and things like that, um, with the downloadable PDFs they were making, but, like, it mattered and, you know, the user was really involved in laying out the steps and one's ability to prompt, well, greatly impacted and influenced the quality of the output.
[00:17:31] Now all of that friction is, Is basically gone and it's simply about telling the robots the outcome that you want and then it's letting it do its thing.
[00:17:39] Right. Uh, you know, is that a, a good thing or a bad thing? I will just say that it's something worth noting.
[00:17:49] All right, uh, again, I have no plan on trying Dots, largely because I don't pay for ChatGPT. I'm not going to, but we will see if Anthropic comes out with their own version and then I will make a decision about that.
[00:18:07] All right, last part. Last year, how I used AI this week. So if you're new here, welcome. Each week I share a quick example of how I use AI that week. This week I used, uh, or rather my use case is in my opinion most notable for what I use and how I use it, as opposed to like, it being some super creative or like, techy use case. So this time I was simply looking for a password that I had forgotten, um, for a website that one of my clients had shared with me and I couldn't find it and I thought that she might have shared the password in an email.
[00:18:42] Uh, and so I went to Gemini inside of Gmail and I searched for it and I didn't find the password because it wasn't in an email. It was actually, um, in a different messaging platform that we use. But it was very nice to be able to just use natural language and explain what I was searching for instead of trying to like, think of the exact perfect keywords to match.
[00:19:06] Um, and so this is very much, very much the same way that dots and museums just works, right? It is just tell it the outcome, use natural language. It'll go and do the thing.
[00:19:18] Um, and so yes, I, I absolutely see the utility. Like in this case, yes, it is more, it's not agentic because it's just like search. It's not like go and send something or build something, but the ability to explain the outcome in natural language and you just have it go do something. You have to go work inside of, um, your, your stuff, your files, things like that. Like that is very, you know, know, dots and muse. And um, I see the utility. I absolutely see the utility. I just, I just don't have a zillion use cases in mind that have me chomping at the bit to use either of those New east, these newest agents. So that is my, those are my thoughts. That is my, uh, my explainer of open AIs dots. Have you used it? Are you using it? I have a client who, uh, is using Metamuse. And I was like, that's cool. Um, but I'd love to hear from you. I use neither one of them. What are your thoughts. Uh, who wants to be my, uh, my ambassador for each of these? Because probably won't be getting into them. Um, but would love to hear from you. So that is all for today. Hopefully you found this episode helpful. If you did, consider leaving a little rating or a little review.
[00:20:30] And don't forget, I have a companion newsletter and blog called the Curious Companion that drops every Thursday. And it's basically the podcast episode in text format.
[00:20:40] So if you prefer to read or you just want a written record of the things, join the newsletter or check out the blog. The little images that I generate, I do use chat GPT for those. Or actually this one, this episode, I use Gemini, um, for it. Um, but check out the little, little images, the feature image. I work hard on those. I really enjoy them. Um, but if you want the written record, join the newsletter, check out the blog. You can head to prompting curiosity.com forward/newsletter or forward slash blog. Or keep it simple and just check out the links in the show notes.
[00:21:16] All right, that's it. As always, endlessly, endlessly, endlessly appreciative for every single one of you. Until we chat again next Thursday, stay curious.
