Name: New in ChatGPT for Work
Uploaded: 2025-03-18T18:46:02.274Z
Duration: 30 min 4 s
Description: New in ChatGPT for Work

Transcript for "New in ChatGPT for Work": Hello, and welcome to our first installment of new In ChatchuBT for work. I'm Puja, a member of our product team. Joining me are Dee and Kian from our solutions engineering team who have built some fun demos to showcase our latest work. We'll start with a quick overview of what we've recently launched, then dive into demos to show you how ChatGPT BT is becoming more interactive, customizable to the way that you work, and capable of taking on tasks independently. Our demos today will zoom into the product so you can see exactly what we're doing. We want a demo at every chance we get, so our promise is that we're going to get into the product in the next few minutes. As you know, over the past year, we've shipped new search capabilities, pair with AI, advanced voice mode, the desktop app, chatbot, it, ChatGPT has evolved from a chatbot back in 2023 to a multimodal assistant and this year into an independent problem solver. As a result, we've been launching features to evolve ChatGPT to adapt to this new world of AI agents. Today, we'll show you how ChatGPT is becoming more interactive with Canvas and Vision and Voice Mode, becoming more customized to the way you work with better context in projects, and by pushing code into your IDE with Work With Apps, and finally becoming more independent with reasoning models like o one Pro Mode and o three Mini and tools like Deep Research. Along the way, we're gonna share some use case examples for different teams and some sample prompts for each of our demos. We'll share these with you at the end of the session as a takeaway. Sounds like it's gonna be a lot of demoing. It is, so we better get into it. As I mentioned before, we've been focusing on making ChatGPT capable of interacting with your team just like you would with a colleague. It can now hear, see live video, and understand documents, images, and graphs. Recently, we also made ChatGPT more interactive interactive with Canvas, an interactive space where you can collaborate on writing and coding with ChatChippity in real time. Dee, do you want to demo some of these interactive capabilities for us? Yeah, Canvas is an interactive space where you can collaborate with ChatChippity. To access, click on the toolbar here and then the Canvas icon. In this case, I'm going to ask ChatChiBT to summarize the transcript from a webinar that we ran last year. Just to give you a look at what I'm uploading, this is the transcript. It's pretty dense right now, but ChatGPT is going to summarize it into a digestible format for us. A new window will pop up when you're in Canvas. This is the space where you'll interact line by line with ChatGPT. We've trained GPT four point zero to be a creative partner that's going to excel in coding and writing. I can collaborate in this space with ChatGPT by making changes and edits directly within the canvas. I can come through here, edit titles, making it a collaborative space. Next, I want ChatGPT to turn this summary into a first draft of an ebook while adding in recommendations and improvements where possible. Something like this would normally take me quite a while. I would need to read up on the transcript. I would need to sort it into sections. I would then need to rewrite it. But here, I get the first draft in a matter of seconds. In Canvas, we can ask questions on specific sections of the text. So I'm gonna ask for this section here. Can I get some more statistics and maybe some consulting articles that I could potentially add into this in the second draft? Now on the left hand sidebar here, you can see search is going off and it's searching the web, consolidating the data, and even coming through and providing some citations for me. When we hover over them, we can see exactly which parts and what, websites it's referencing. Canvas also has the ability to write shortcuts to expedite your editing. So you can edit directly and ask for edit suggestions. You can ask for just the length, reading level, final polish, or even some fun emojis. What we're really most excited about, is the coding functionality of Canvas. Let me show you what I mean by that. Say I have a PRD and the key capabilities and features for a new product that Acme Corp is working on. I wanna be able to quickly build a prototype to share this with my team members. So here's a look at the the PRD that I would need to build into my prototype. So what I'm gonna do is I'm gonna ask ChatGPT in plain language to build this prototype based on this uploaded PRD. Canvas has the ability to leverage o one, so I'm gonna build a really well thought out comprehensive code for this. So what we're gonna see here is Canvas is leveraging o one's chain of thought to plan out this work for me. And now you can see the code written here in Canvas. So despite me not really knowing how to code or where I should even start with this, I can mock up a prototype really quickly. Like before, you can edit code directly within Canvas and ask questions about certain, lines of code here. Using the shortcuts bar, you can also get inline suggestions, ask for improvement, port, debug, and even add in logs and comments. What were recently released, though, is code execution for HTML and React. So which means I can go over here and click on preview and the visualizations are going to appear. And I can even interact with this. So right now, I have a really quick prototype of my unified, analytics dashboard. I can come up here and select different roles. What would marketing be interested in? Or what if I was finance? And we can see that this is changing live in front of me. I can even ask Puja here. It looks like our marketing automation has an error. Maybe you can look into it for us. Really, it is just a more interactive environment so you can show this to team members to really bring your vision to life. Canvas runs in a sandbox environment, so I can make API calls with network controls for security, of course. Puja, do you wanna take over and walk us through some of the best use cases for Canvas? Absolutely. Canvas helps teams iterate and collaborate to be more creative, augment their skill sets, and polish their work. Marketing teams are using Canvas for content generation and interactive data visualization. As you just saw, product teams use Canvas for writing PRDs and prototyping ideas without needing to code. And L and D teams are using Canvas to build new interactive learning experiences. And from a security and compliance perspective, we've included network access controls and logging in the compliance API for Canvas. Chat GPT has also become more powerful on your phone. We recently launched vision in advanced voice mode, meaning your teams can now use their mobile camera to interact and understand their surroundings, screen share for more context with ChatGPT conversations on the go. Kyung, can you walk us through how that works? Sure. Advanced voice mode is a crowd favorite, but now that it has vision, we think it's even more powerful. To open and vision in ChatGPT, enable advanced voice mode first. Next, you can hit the camera screen, and then it will have access to your surroundings. For example, I might ask Chachi Bitti about this book. Hey, Chachi. What's a good baking recipe for a croissant? What should I make? Hey. A classic croissant recipe is a great choice. It's a bit time consuming, but totally worth it. You'll need flour, yeast, sugar, salt, butter, and milk. The key is to fold the dough multiple times to create those flaky layers. Are you ready to give it a go? I can even ask it where I could buy this book online. Sounds delicious. Can you search online where I can buy this book for my friend? You can buy the Tartein Cookbook online through retailers like Amazon, Barnes and Noble, or directly from Chronicle Books. When I go back into the chat, you can see the conversation I just had in the text and all of the sources. I can also share the screen on the go. Let's say I'm on my way to a client meeting and you just received a email from your client asking for a discussion around a topic you aren't well prepared for. The email contains a complex cohort analysis. Let's see how I can ask ChattGPT help. I'll start off by activating advanced voice mode again. I'll hit the three dots here, share screen, start broadcast. Now ChattGPT can access the screen, and I can share what I want to with it. Hey, Chat. Pull some insights from this chart. What revenue trends do you see for the January cohort? Hey. So looking at the cohort analysis chart, the January cohort shows some interesting trends. They started with a higher initial revenue, which gradually decreased over the months. It looks like their revenue contribution tapers off after a few months. Anything specific you want to dive into? Yeah. What cohort did the best and should we try to replicate? The March cohort seems to have performed the best with more consistent revenue over time. Replicating whatever strategies or campaigns were used then could be useful. Any specific areas you want to explore to replicate that success? And just like that, ChatGPT is helping me prepare for my meeting life on the go. Thanks, Kyung. Vision and voice is perfect for busy executives who need a personal assistant to review emails, calendars, or even Slack messages on the go, or for understanding products or equipment in person. Vision and voice and Canvas are moving ChatGPT towards a future where it can collaborate side by side with you and your team in a richer, more natural way, leading to faster adoption, more exciting and complex use cases, and better results from AI human collaboration. In addition to making ChatGPT more interactive, we've also focused on customizing ChatGPT to your workflows and tech stack. To do do this, we've improved ChatGPT's ability to remember preferences, from the way you write to the depth of answers you want. It can understand your business context and connect to tools in your tech stack. Let's start with how we're helping you provide more context to ChatGPT. We launched Projects in ChatGPT in January. Projects let you organize charts into folders and use uploaded files as context for conversations within a project. Dee, do you want to walk us through how that works? Absolutely. Projects live in the sidebar above your chats. What we're going to do is we're going to create a new project here today by clicking on our sidebar, new project, and creating a new name. We're going to look at this as an analytic product launch. Now we want our project to have some context. In this case, I'm going to upload a PRD, some, customer feature request, and maybe a market analysis conducted via deep research. Once I've uploaded a document, it becomes context for any of our chats within this project. You can also set up some custom instructions to tailor away that ChatGPT responds. In this case, I want Chat to pretend to be a product operations advisor or VP of product. Now any chat within projects has this context, and it also has the instructions. I'm looking to build out a detailed launch plan. I want to look into different regions and to find any potential risks. Projects works with o one, and I'm going to select this because I want to build a complex launch plan taking into consideration a bunch of constraints. As you can see with o one, it has a chain of thought when crafting my launch plan. It's referencing the documents I've uploaded. It's looking at the marketing and engineering teams. It's really giving me a really robust answer. We can now see that the output is a pretty detailed six week launch plan. Normally, putting a first draft together like this would mean slowly writing and referencing all my uploaded documents, but here, I'm able to have a first draft in a number of minutes. Relevant documents are also referenced throughout this showing that they've been read by Chat GbT before building in the plan. You can see here that we've referenced a few of them. So let's say I want to want to find events that we want our marketing team to present at a launch in North America. I'm going to start a new chat within our project and then I'm going to add in my prompt to say that we want to be specifically focusing on North America and Europe. I'm going to enable four o so that I have my search capabilities. As you can see here, we're going out and searching, so it's going through the web to find out what are some of the relevant, events within these locations. We can now see that it has shareable links and citations so that we can follow-up on this later on. We can now return to our homepage where we're gonna have the history of the two chats that we've asked within our projects. It's a really powerful way to centralize chats in a particular work stream and ensure ChatGPT has the same context as your project team to help you work smarter and faster. Thanks, Dee. Projects are valuable whenever you wanna centralize all your chats on a particular work stream, giving ChatGPT the same context as any of your colleagues might have. Marketing teams can use projects to upload launch plans, messaging source documents, and audience research exec review notes to improve product launches. And sales teams can upload relevant research, meeting notes, and account plans to better understand their prospects. We know many of our customers juggle multiple SaaS tools and often ChatGPT is right by their side. But up until very recently, you could only copy and paste between tools. In our Mac OS app, ChatGPT can understand full context within some IDEs Not just what's on your screen, but it can now even edit information in those apps when enabled for your workspace Kyung, you're up. Can you show us how that works? Sure thing. We're gradually adding native app integrations. You can open ChatGPT with the option plus space bar. When you click the work with apps button, you'll see a list of available apps that ChatGPT can connect with. To enable this, you'll need to grant permissions either in your system settings or directly within the app, depending on which app you want to enable. Once it's set up, ChatGPT still won't have access to any app until you explicitly select it. Once it's connected, it can pull context from within the app, not just what's visible on the screen. For example, as a developer, I might share the code on my IDE. Let's take a look at how this might work as a developer. I'm working on a checkup page for my new bookstore. However, when I put in my payment information, I hit an error. The error shows that I have an issue with my integration with my payment provider, and it looks like I'm missing a parameter. I'll open up ChatGPT via the option plus space bar shortcut and ask it for help fixing the error. I also know that this error is specifically coming from my checkout page, so I'll click the work with apps button, which shares the entire code from my four print file in my IDE with ChatGPT. In addition, I'll swap out my model to use o three MiniHi. It's a little bit slower, but works really well with coding use cases. You can see ChatGPT go go to work starting off with the chain of thought. You have full visibility into the steps the model is taking to come up with a solution. And here you can see the progress and even have ChatGPT automatically edit your file with your within your IDE if you'd like. For this example, let's review the file together first. It looks like I used camel case instead of snake case, which the payment provider requires. So let's go ahead and push the changes. And just like that, you can see the changes reflected in my IDE with the return URL being updated to snake snake case. Let's save the code and give that another try. Great. Problem solved. So previously, I would have had to copy and paste my, paste my error, perhaps look up the API docs, but now I can share my code directly with ChatGPT and ask it to update it as well. Engineers love work with apps for bug bashing, code rewrites, searching for APIs, writing test cases, and just generally getting up to speed on the code in front of them. Meanwhile, non engineering teams use it with apps like Notion for faster content creation, especially by dictating ideas in voice mode and refining them with ChatGPT's help. In short, we're helping ChatGPT adapt to your tech stack so that it can show up whenever you need it, and we're giving you the tools to build a more tailored contextual experience that leads to work that meets your company's standards. Now that we've talked about how ChatGPT has become more interactive and customizable, let's talk about how ChatGPT is becoming more capable of taking on some tasks independently. With three recent launches, o one Pro Mode, o o3 Mini, and Deep Research, you can start to see the foundations of a truly agentic tool Let's begin with o3 Mini and o1 Pro Mode o3 Mini and o1 Pro Mode take time to think, apply logic, and handle multi step tasks. They excel at math, science, coding, and complex problem solving. O three minutei is lightweight and efficient while o one Pro mode may take a few minutes to deeply reason through an answer. Dee, can you show us how o one pro mode works? O one pro mode is a reasoning model, which means it takes more time to think through its answers, it can plan, and it excels at logical thinking. The best way to show it off is to look at its chain of thought. So let's pretend I'm a corporate financer at a sneaker company, and I need to research feasibility and opportunities within the Japanese market. I'm gonna select o one pro mode from our model pickle picker and drop in a bunch of questions that we'll need answered before our expansion. You could see here I'm adding a lot of complexity here to show how powerful o one pro mode is. O one pro mode will also take longer to think even than o three minutei, but both will show you their chain of thought. As it works, you'll see the chain of thought displayed, showing how it reasons and plans through tasks and applies logic. You can see here it's analyzing the market, it's looking at consumer trends, it's determining key brands, and considering marketing strategies in this region to work. The difference between these models and our foundational models like four o or 4.5, is that these models actually take time to work through a multi step problem. In this case, it can take around three minutes. So rather it than instantly responding, it really does take that next step. In contrast, four o and 4.5 will be faster and more intuitive and better at natural conversations. So as we can see here, it took about three minutes, and we're able to see a really detailed report that goes into market sizing, consumer segment, key players, and really well thought out answers to all of our questions. A strategy document like this would normally take a long time to build with multiple people, even as a first draft. But with our o one Pro mode, we have a great start to the analysis in minutes. O one Pro mode in o three minutei's extra time spent thinking means it can tackle complicated planning, strategic problem solving, and help with engineering and math problems. Here is a list of sample use cases. And that leads us to deep research, our first truly agentic tool. To build it, we trained our yet to be released o three model to plan and reason through detailed research, search the web, and use Python to analyze charts and graphs essentially teaching you to use multiple tools The result is an agent that can do multi step research across dozens of sources compiling detailed reports with inline citations Can, can you give us a tour? Deep Research is our first agentic tool. You can click Deep Research in the toolbar for access. I've pasted in a prompt asking ChatGPT to help me with an expansion strategy as I'm looking to expand my D2C sneaker store internationally. I need help prioritizing whether to launch in Japan, Korea, or Taiwan First. I've given it some criteria around what I want it to take into consideration and how I want the final structure report structured. I can also guide it with types of sources I want it to use. Now let's send that off. Now you can see ChatGPT will start off by asking me with some clarifying questions, just like any good researcher would do. I'll provide it with some more details, but I'm also gonna leave it a little open ended in some areas, and I'll ask ChatGPT to make recommendations. ChatGPT returns with a plan of action based on your original prompt and answers to its follow-up questions so that you have visibility into what it's doing. Once we send it, it can take anywhere from five to thirty minutes. As the research begins, you can see in the chain of thought on the right hand side how the model is scanning the relevant articles on the web, providing a completed run that I ran earlier today. Let's take a look at that one. This research took us about eight minutes. When I click here, I can see the full chain of thought from the run and all of the sources that it access. What's amazing is the depth and the detail in the report. In this report, you'll see that ChatGPT has cited all of its sources so that I can navigate to have a closer look if I wanted to. It's aggregated the total addressable market, the consumer trends, the the competitive insights, and the expansion strategy for every single country that I asked it about, and references them all with inline citations, which again allows you to verify sources. Along with the detailed research report, we also have a summarized view in a table format, which serves as a nice executive summary. This is the kind of research that normally would have taken me hours or days, but deep research can do it in a fraction of time by searching the web, analyzing documents and graphs, and shifting its focus as needed. Puja, how are other customers using it? Great question. Teams across the board are loving deep research. Sales uses it for account and industry insights, marketing for researching new audiences or industry trends and data points, and product for competitive research and feature planning. To recap, we've been busy improving ChatGPT as it evolves from a multimodal assistant to being capable of solving problems independently. Canvas, Envision, and Voice make ChatGPT far more interactive and natural. Projects and work with apps provide a more personalized experience that has all the context that a coworker might have about the way your teams work. And o one Pro mode and o three minutei along with deep research move ChatGPT closer to a truly adjunctive tool that can take on multi step tasks so your employees can focus on other work. And if you wanna get started sharing these use cases with your team, we have two separate plans. Our team plan is best for companies that wanna get started quickly with the secure and collaborative version of ChatGPT. For large organizations who have more complex compliance and security requirements, our enterprise plan is best. We encourage you to try these new features now. Live attendees will receive prompt and use case packs at the end of this webinar and will send out a survey to gather feedback about future sessions you'd like to see. We've also shared help docs and security guidance for all the features we shared today in this webinar. Thank you for joining us and we'll see you next time. Bye. Bye.