ChatGPT Branching, Project Memory & More: Top 5 Hidden Features. #chatgpt

Today I want to share 5 big updates in ChatGPT that really only power users know how to use. And it's hard to keep track of all the updates that come, and some of them are meaningful, some of them are not. Some of them get overwritten by ChatGPT in later versions. But these are 5 big ones that have happened in the last several months that if you really want to up skill yourself and become a power user of ChatGPT, you need to know about these. These really are game changers if you want to squeeze out more intelligence, more speed, more insight from your AI workflows. And they're going to affect the strategy that you use in interacting with ChatGPT at work. And the goal here, as you know if you've watched my channel— and if you haven't, go ahead and subscribe and like, share, The goal is to help you really leverage AI at work. We know that AI is coming. We know that it's going to affect all of our jobs. We know for certain that if you know how to use these AI tools successfully, that you are going to become amplified. You're going to become better at whatever it is that you do, and you're going to be able to compete against other companies and co-workers when the squeeze comes, when it becomes clear that we only want the best and brightest and best power users of AI within a particular organization. So the first one I want to talk about is this new branching feature. It allows you to branch conversations within a thread, and you can fork Any part of that thread into a new chat thread. This allows you to explore different approaches. Okay, I got this particular thread going. I want to now branch out and start this into a new thread. I'm going to keep going in this thread, and now I've got 2 competing threads that I can interact with in different ways. Maybe as a lawyer, I might say, help me argue on behalf of my client. In the other thread, you're helping me argue against my client so you can understand the adversary's point of view. There are Literally dozens of different ways why being able to branch a thread into a new thread while preserving the memory of that thread is going to be super helpful. Think about it this way. You are interacting with intelligence. You're not having it just draft for you. You're trying to learn, you're trying to explore, you're trying to brainstorm. These lines of reasoning that you tend to go down as you're engaging with a particular chat thread, end up getting baked into the thread. And if you start out in one direction where you're guiding the AI saying, here's what I'm thinking and what I really want to explore is this. And by the way, my overall approach is going to be down this path. I want you to really consider these particular things. Here's what I think about the issue or the problem. You should be talking to your AI, by the way. You should be talking to your AI that way. You should be exploring But once you start down in a thread, down a particular line where you've given it your thoughts, your biases, your input, once you go down the thread, you can't unlearn that if you're the AI tool. You can tell the AI thread, and this is the way we used to do it, we tell the thread, hey, look, now I want you to go in the other direction. I want you to think of the counterargument. Okay, but remember that you're dealing with a token chain. And the first input into the thread is going into the token chain. The knowledge you upload is going into the token chain. This custom instructions is going into the token stream. Then every input and output is going into that token chain, and all of that is going to be considered as you maneuver through the thread. So this token chain of input/output knowledge and constructions, and perhaps even some of the custom instructions you provided at a system level. These are all going to be part of that token chain. In the world of AI, less is often more, meaning if you're trying to really explore and go deep into a very narrow issue, or you're trying to push in a certain direction, then excluding information from the token chain becomes as important or more important than including information in the token chain. So again, by allowing you to explore alternate threads and fork a thread, it allows you to explore different lines of reasoning, consider different options side by side. One option A, one option B, which one looks better? And you are able to do this without losing your original conversation. Without losing your original conversation, what does this mean? It means that if you introduce new tokens into the chain, upload a document, or upload your thoughts, right? All of that is really good stuff, and you might not want to corrupt it, right? You want to keep that conversation going potentially while going in a new direction. So you get to keep that original conversation without contaminating it with new thoughts, new strategies, new knowledge upload. Really, really important. And if you're just working within a single thread and you're going and you're going and going forever, that is one of 1,000 different reasons why you may be using AI wrong or inefficiently or failing to get the most out of your AI tool. Remember, we talk a lot about the 4 pillars of AI, which is you need the right skills, you need the right strategy, right? You need the right mindset, and then you need to measure the results. So strategy and skills are really important. Your strategy and how you go about interacting with that thread is where you are going to get from 5x to 10x, 10x to 100x. Really need to be always thinking about how you're interacting with the AI and how to get more out of it. So if you're evaluating arguments or you're evaluating strategies or trying to explore different legal approaches in my business law, this lets you test multiple versions of Perello without polluting the original thread. So that is a really important new feature within ChatGPT that didn't get a lot of press but is really critical. The second one is project-only memory. And if you've seen some of my training videos which show you how to set up project folders as a kind of a next level of GPT assistance— remember, we had GPT assistance that allowed you to upload knowledge and custom instructions into a particular assistant at a particular job. Then I think ChatGPT realized as a result of, you know, users using, you know, GPT assistants, that there was a better way. And that better way became the ability to create project folders within ChatGPT so that the special knowledge that you upload into that— now not your thread, but that folder— and the custom instructions for that folder apply to all the threads. And that you can have a thread for meta-prompting, getting a better prompt instruction. You can have a thread for analyzing a particular document, give it a good label so you can come back to it. For drafting, you can have all these different threads, and it's all going to be controlled at a project level by those instructions, by the uploaded knowledge. That was a big leap forward on GPT assistants. Well, now they've got project-only memory, and for heavy or sensitive work, Power users can now enable memory that is confined to a project. Now, as a lawyer, what that means is I got a client. I'm going to create a project folder for that client. I am going to upload specific documents into the knowledge for that client. I'm going to give it instructions to help me as a lawyer assist that client. And now I can toggle and say to ChatGPT, but I only want you to have memory For this particular project folder, for this particular client. Before, if you said to ChatGPT, hey, I want you to remember this for future, it would remember it and it put it somewhere where it could be applied to all your threads. This is a whole leap forward because now you can say, I want you to remember these things for this client, for this project, for this folder, and it will remember them only for that folder. So it really allows you to add Again, to that token chain, relevant information that is going to need to be repurposed. So let me give you an example. So today I'm using ChatGPT to draft a trial brief, and there are certain themes and certain winning arguments that I want to bake into all the work product across all the threads. And so I tell ChatGPT, hey, look at— here's what I think is important. Here's what I think wins the case. Here's the overarching theme that we need to have permeate everything. And then you have to use the magic words. I want you to remember this, put it into memory. I want you to remember this, and then it will store it for you within that project folder. So this means that ChatGPT uses context from potentially other chats. You could say, remember this whole thread. And now when you go to a new thread, it's going to remember some important information, and it's going to only be within that project. And it doesn't carry into unrelated context or unrelated folders or unrelated threads. that have nothing to do with that particular project folder. So what happens inside the project won't leak out into other chats. So this is especially useful for long cases and ongoing research and cross-department work in business or in law. So that is going to be key, right? What we're, what we're doing is we're feeding, you know, the model was trained on a certain dataset. Now we're adding to that model certain contextual information that's going to be, that's going to be part of the intelligence that is going to be spit out to you in the output. Being able to say, I want you to remember this for future threads and have it exist within a project folder is going to allow you to get much better output from the threads within that project folder. A lot of people don't even know this exists. A lot of people just get in there and chat and start a new thread every once in a while. That is beginner level zero. The nuances of how you can get the most out of these tools as you start to interact with infinite potential intelligence is going to be found in the skills, the strategies, the mindset that you bring to your desk in the morning when you start interacting. There is no upper limit On the amount of intelligence and amplification that you can realize. The only limit is you. It's not the tool. It is only your limited skills, your limited mindset, your limited methodology, your limited strategies. So as you expand all of the different skills that you have, as you bring a better mindset, better understanding to the AI interaction, and as you develop better strategies for interacting with a particular thread, with a particular project folder, with the AI in general, that's where your amplification's going to occur. Number 3, new feature that you probably didn't know exists is one that was kind of part of the prior iterations of ChatGPT and now has been brought back into ChatGPT-5, and it's the model selector. So there's all these modes that have been in ChatGPT. ChatGPT-5 was supposed to say, oh, it's a single mode. ChatGPT-5, and it'll figure out whether or not to use reasoning or to go fast or go slow or to do research. It'll figure it out for you. Well, people didn't like that, and maybe they didn't like it for a good reason. Maybe it just didn't work as well as people wanted it to or as intended. But now, even within the new ChatGPT-5, you can pick some of the older models if you want, and there are several models of ChatGPT-5 that you can Select. This is part of your strategy. What are you doing? Are you just asking for output analysis of text into more text, or are you needed to really think about a problem? Do you need to do independent research? Based on these issues, you're going to select the right model, and by selecting the right model, you're going to get further amplification. There are modes. The default mode is auto. which is it's going to select for you. Then there's fast, which is going to be a faster model, which is not going to spend a lot of time. Where do I, you know, where do I go to dinner tonight? Then there's going to be thinking, where you want it to do some deep, deep thinking about a problem and potentially even deep reasoning as an option. So power users choose the mode appropriate to the task. And for now, if you know how, to think about which one to choose, you're going to be much better off. Use Thinking for deep reasoning and complex legal analysis and high-stakes drafting. Use Faster Auto for rapid responses when you're looking to move fast. You only need a summary. It's just looking at text and spitting out text. You can also click on the Show Additional Models and go back to GPT-4, which was more empathy, more personal touch, or O3, which did really good analytical work. You can compare O3 with 5 deep reasoning model and see which one's the better result. You get a second opinion when you go to the doctor. You might want a second opinion when you use AI. So these are really important things to know exist and to learn how to use. If you're just sitting down and using ChatGPT, you don't know these different selections are available and you're not selecting them appropriately, really, you're going to get amplified, but not near as much as I'm getting amplified. All right, number 4. Okay, number 4 is a big one as well, and you're going to be surprised, I think a lot of you, that this exists. But there are now connectors within ChatGPT where I can connect ChatGPT directly to my Outlook email, my Outlook calendar, Microsoft Teams, in the Google world, Gmail, Contacts, Google Drive. You can connect directly to those tools within ChatGPT. So we're kind of starting to move in this direction where we're getting beyond cut and paste. Well, I want to go cut and paste this Teams document or download the Teams document, Word document, upload it into ChatGPT. Now we have this kind of first step towards saying, well, I'm just going to connect it directly to my, to my Teams channel. I'm going to connect it directly to my Outlook email and say, hey, I want to know what I've missed over the last week that may have fallen through the cracks. Do a deep dive analysis in last week's worth of emails. And then I'm going to select it to connect to Outlook within ChatGPT chat. And it's now going to have access to this. So if you haven't tried this one, this one is a pretty substantial change to how AI is going to fit into your daily job. It is the first of what will become many steps where ChatGPT integrates into everything without having to connect it, download, upload, these types of things. These connectors are going to only improve over time. And this AI connection ability is going to allow you to access emails and meetings and schedules and documents and the rest. It can also, you know, hit Dropbox and OneDrive or Box. So it There's a number of things it can connect to. That list of things that it can connect to is only going to increase. Power users use this to kind of build workflows that reduce friction. And what I'm finding is, again, this is kind of version 1 stuff, right? But you want to start to learn how to use this because they're going to make it better over time. So I will sometimes start the day by connecting to Outlook email and asking questions about A specific client. So instead of having to search in Outlook, then look at each email, I can just say, hey, look, this particular client, I've been communicating with them over the last 2 weeks. Here's their name or here's their email. I want you to do a summary of the conversations that we've had, identify the issues that we've talked about, and then tell me what you think I still need to address that hasn't been addressed for the client. If you were the client, what might you be expecting from me? Perfect. Number 5, this is the last one for today, right? And again, go ahead and subscribe to this channel. There's tons of content that just goes back for almost 2 and a half years as I've been training, uh, business professionals to use AI tools. And those videos, by the way, I get comments all the time, I can't believe this from 2 years ago is still exactly relevant and on point to where I am in my AI journey. So number 5 is voice and tone improvements. It is an advancement to standard voice modes where— and personalities, et cetera, all the different things that you can do with voice mode. And voice mode is where you just talk to the AI and it talks back to you in a voice of your choice. ChatGPT's voice mode has become more refined in the last 6 months. There's more natural inflection. It's more conversational. You can interrupt it. It's got a better kind of tone, emotional tone to that.

Good afternoon. I want to tell people, business professionals, about voice mode and why they should use it. Give me your best reasons.

Sure thing. So for business professionals, voice mode is a pretty handy tool because it makes communication a lot more natural and efficient. Instead of typing everything out, you can just speak your thoughts, which saves you time and often gets your point across more clearly. It also helps you multitask and keeps things feeling a bit more personal and engaging when you're working with colleagues or clients. So it's just a nice way to streamline how you communicate.

And on top of it, there are now these personality settings. So you can tweak the style or persona of ChatGPT to be warmer, more familiar, more professional, more antagonistic. And power users can adjust the tone when they are working with voice mode. I'm guaranteeing you that there's very few of you who are using Voice Mode regularly. I use it, and I'm using it more and more in lots of different use cases. And I did a video the other day, was driving to a court hearing. I uploaded the briefs from both parties. I uploaded the key documents into a thread, and then I enabled Voice Mode. I told Voice Mode what I was doing. I was going to argue this motion. I was going to want it to spar with me and help me hone the arguments. Gave it that instruction to adjust the style and voice of my conversation. And then I drove and I talked the whole way and it worked with me as though I had this expert moot court coach in the car with me helping prepare for the argument, which we in fact won. So you will start to see more Cool functionality in Voice Mode. I encourage you to get in and start using it for a variety of different things when you're not at your keyboard, right? Not at your keyboard, you can still interact with AI. You can dictate into it, it transcribes it, it gives you output. You got to look at it. Hey, Voice Mode solves all that. And I, I'll tell you a secret. AI is going towards Voice Mode. We are not going to have keyboards. We are not going to have mouses. Where we're going to be interacting with AI 12 months from now, 24 months from now, primarily by having these conversations and telling these agents that are already being developed what to do. Hey, go adjust the brief. I like this. Insert it into the brief. You won't be typing it. You won't be cutting and pasting it. So check out voice mode. And remember, we have 5 new items here that are really important. Branching or forking a thread, number 1. Project-only memory, number 2. Model selection, picking the right model for the job at hand, number 3. 4 connectors where you can connect your Thread to Google tools or Microsoft tools in order to be able to work off of a file, a cloud-based system in order to access that information from directly within the chat. And then the last one, of course, is voice mode improvements and advancements. My name is Enrico Schaefer. We'll see you next time right here on this channel. Subscribe, share, like, and feel free to leave me comments, ask questions. I answer all of them. Until next time.