Ep 421: OpenAI’s new ‘12 days’ features: What’s here so far and what it means

Resources:

Join the discussion: Ask Jordan questions on OpenAI's new updates


Upcoming Episodes: Check out the upcoming Everyday AI Livestream lineup

Connect with Jordan Wilson: LinkedIn Profile

Try Our Free AI Prompting Course: Register for our free Prime, Prompt, and Polish AI Course! 

OpenAI's 12 Days Features and Google's AI Breakthroughs

The tech space is ever-evolving, and with OpenAI's latest '12 days' feature roll-out and Google's impressive AI advancements, it seems AI's future is brighter than ever. These aes result in monumental shifts, opening doors for a range of businesses to leverage AI technologies for growth and efficiency.

OpenAI Takes the Lead with Reinforcement Fine Tuning

OpenAI's new Reinforcement Fine Tuning Research Program (RFTRP) is set to revolutionize custom AI model development. Using reinforcement learning, RFTRP enables the creation of specialized models in narrow domains like medical or legal with minimal data, resulting in enhanced model reasoning and accuracy. The program promises to be a game-changer for smaller businesses and non-technical users making AI more accessible.

Canvas: A Comprehensive Tool for ChatGPT Users

In another significant move, OpenAI has decided to make Canvas - a collaborative tool, available for all ChatGPT users. Integrated into the tools section, Canvas serves as an ideal side panel for writing and coding. Its unrivaled features include direct editing of text, version history, real-time document editing and Python code execution. Despite the lack of the full code versatility of Claude's Artifacts feature, Canvas's accessibility sets it apart.

The Rise of Google's AI Evolutions

Google's Gemini 2.0 Flash AI model, offering a multilingual, multimodal interface for text, imagery, and audio inputs, delivers amplified speed and performance. Furthermore, Google's trailblazing projects like Project Jules, Project Mariner, and Project Astra, position it as a front-runner, integrating AI seamlessly into practical applications.

AI Research Made Simple with Deep Research

Google's Deep Research tool, automating online research by scanning widespread content online, is a testament to the company's AI innovation. This tool enables faster, more accurate research, perfect for busy professionals without the extra hours to dive deep into the web.

OpenAI's Premium Plans - A Taste of Future Pricing?

OpenAI's recent unveiling of Chat GPT Pro and O1 model underscore a growing shift towards premium AI offerings. These advanced tools cater expertly to a wide swath of fields, from data and business intelligence to academia. Speculation about future high-priced plans from tech titans like Microsoft and Google just adds another layer to the intrigue.

Canvas Integration: A Boon for Collaboration

Canvas' integration with ChatGPT opens up an innovative collaborative platform for simultaneous writing and coding tasks. With features including inline feedback, targeted editing and one-click modifications, Canvas fosters a robust context-awareness environment beneficial for both free and paid users.

Apple and ChatGPT: A Match Made for Efficiency

Apple's recent integration of ChatGPT into its functions promises enhanced reasoning, advanced content generation and image analysis. However, one has to note the branding of AI as "Apple Intelligence", a move that puts the company slightly behind its competitors.

OpenAI's Sora - A New Dawn for Video Experience

OpenAI's Sora model can transform simple text prompts into high-definition videos. While not perfect - real-world physics offer a particular challenge - an understanding of Sora's capabilities allows for the creation of professional-grade content at a fraction of traditional costs. However, limited availability and issues with server reliability indicate potential areas of growth for this technology.

Looking Forward in the AI Space

Following the pattern of announcements and releases, it is likely to observe AI's gallop to peak efficiency. Google and OpenAI have laid down the gauntlet, and it's up to the enterprise users to make the most of these advancements. Indeed, AI's future is happening now, and it's an exciting reality all business owners and decision-makers should be ready to embrace.

• Topics Covered in This Episode

1. Expansion of OpenAI's Reinforcement Fine Tuning Research Program
2. Updates to Canvas in OpenAI's ChatGPT
3. Google's AI advancements including Gemini 2.0 Flash
4. Google's deep research tool and its market implications
5. Day 1 of OpenAI's 12-day feature release event
6. O1 model and Chat GPT Pro announcement
7. Predictions on further premium features in AI plans
8. Canvas and ChatGPT integration
9. Siri and ChatGPT integration in Apple's devices
10. Discussion on Sora, OpenAI's text-to-video model
11. Limitations and the future of Sora's availability
12. Account tiers and upgrade limitations in OpenAI's offering
13. Competitive landscape: OpenAI, Google, and Gemini
14. Opinion on Google's market strategy
15. Personal user experience with Google's deep research tool
16. Predictions about OpenAI's upcoming releases

Podcast Transcript

Jordan Wilson [00:00:02]:
If you want new AI features, this is the show for you. My gosh. So, today, we're gonna be going over what's new and what's been announced with OpenAI's new, updates. They've had this 12 days of ship miss or 12 days of OpenAI, and we're just about in the middle of it. So today, we're gonna be recapping not just what's new with all of these updates, but what they mean and how you can actually use them. So, I'm excited for today's show. I hope you are too. What's going on y'all? My name is Jordan Wilson, and I'm the host of Everyday AI.

Jordan Wilson [00:00:43]:
So this thing, it's for you. It's your daily livestream, podcast, and free daily newsletter helping everyday people like you and me not just learn about AI, but how we can all actually leverage it to grow our companies and our careers. Is that you? If so, this is for you. Another thing that's for you, our website. My gosh. You gotta go to it. It's your everyday a i.com. There, you can listen to anything.

Jordan Wilson [00:01:10]:
Right? Whatever you wanna learn in generative AI, we have it there. We've talked to the world's leading experts on just about everything. You can go watch, read, and listen every single episode, 420 plus, as well as go sign up for our newsletter where we recap each day's show as well as bring you everything else that you need to know to stay up to date. Alright. So today's show is gonna be a little different, a little special, I think. Right? Because every day we start off with the AI news. And today, actually, Google shipped so much. Today's episode, I mean, it's gonna turn into a Google slash OpenAI update.

Jordan Wilson [00:01:46]:
So, yeah, we're gonna go over, the 12, or at least, you know, the first 5 days of of OpenAI's updates. But, going over the AI news today, well, Google's got it on lock because they announced, I think let me just say this out loud. I think Google announced more in the last 24 hours than OpenAI has in the first 5 days of their, kind of 12 days of OpenAI. So we're gonna start off the AI news, which is literally just Google updates because I think they're that big and that significant. Also, let me know, livestream audience. I had some issues last time. Hopefully, y'all can hear me today. Alright.

Jordan Wilson [00:02:28]:
But let's talk about AI news today. It's just all Google. My gosh. They shipped. Alright. So first, Google has unveiled Gemini 2.0 Flash, the latest in its AI model lineup, promising enhanced speed and performance over other previous versions. So the new model is part of Google's strategy to encourage developers to create innovative applications using AI Studio and Vertex AI. So, yeah, if you're going on the front end of Google Gemini, you're not going to see Gemini 2 point o, Flash.

Jordan Wilson [00:02:59]:
Or, actually, you know what? There's been so many updates, over the last, you know, couple of days. Okay. Actually, I'm wrong. You will see 2 point o flash on the front end. Yeah. That's right. Finally, we get some front end updates for Google Gemini. So Gemini 2 point o flash is a multilingual and multimodal, supporting text imagery and audio inputs and can respond in any of those modes.

Jordan Wilson [00:03:21]:
So the model introduces a, and this is on the Google AI Studio side, a multimodal live API enabling real time conversation and image analysis. So Google aims to integrate AI agents, also into practical applications. Yeah. And we're gonna be talking about some of these, exploring projects like Project Astra, Project Mariner, and Jules, which focuses on human agent interaction and coding assistance. So, Jules, let's talk about that very quickly. That is a new AI powered coding agent, so specifically made for coding, and development tasks, and it's set to assist developers with coding tasks in Google collab and other IDEs through currently avail and that's only available right now to, trusted testers. Alright. The next one, Google's Project Mariner.

Jordan Wilson [00:04:14]:
Alright. So this was code named previously. I believe it was project Jarvis. You know, I said, that name's not gonna happen because that's gonna be another lawsuit like the last, company that tried to do Jarvis. So Google has, officially unveiled the project Mariner, which is the 1st AI agents capable of navigating the web autonomously. So, yeah, you can kind of do it, right now with Claude's computer use, but you can't really do it on the front end. You have to do a little bit of setup, you know, downs, download something on your desktop. So this is the first one.

Jordan Wilson [00:04:46]:
It's not available to everyone yet, but, all these new things that we're talking about here from Google Gemini are powered by the new Gemini 2.0 Flash. So this new Mariner is a Gemini powered AI agent that can control a Chrome browser, moving the cursor, clicking buttons, and filling out forms, mimicking human interactions on website. So it's initially just released to a small group of these preselected testers. So, yeah, who knows? Or, actually, the rest of us are gonna see it. But it's moving toward an AI mediated web interaction rather than just direct user engagement. So the Mariner AI agent can perform tasks such as creating shopping carts on grocery websites, finding flights, and exploring recipes. Here's how it works. So screen activity is monitored through screenshots sent to Google Gemini in the cloud, ensuring users are aware of the agent's actions.

Jordan Wilson [00:05:36]:
And right now, Project Mariner operates only in the active tab in Chrome. So it does require users to kind of watch its activity or when I'm using it, I'm just gonna give it its own computer and be like, yo, go do your thing, buddy. And Google says that is intentional for transparency. There is no specific timeline for a broader rollout of this project mariner, but, you know, we'll see. Google, a lot of times, they either ship something right away or they, you know, put out a little, tease and it could be many, many, many months like we just saw with Sora. Alright. Project Astra, Google. Yes.

Jordan Wilson [00:06:11]:
More Google AI news. Y'all, I wanted to bring you other AI news, but I didn't have a choice. So, now we have, a wait list. Yay. Google has opened project Astra, which we've talked about many times on the show, to its trusted tester wait list. So this is its new augmented reality, kind of project, very similar to what we saw demoed on 60 minutes from OpenAI with its advanced voice mode, the ability to kind of see real time video and interact with that. So a little different than some of these other things that we're talking about. But according to Google's DeepMind page, Project Astra allows users to engage via a mobile device and prototype AR glasses aiming to enhance user interaction through an experimental Astra app.

Jordan Wilson [00:06:55]:
So participants can use the app to, to open their camera and share their screen, gaining quick information about objects they see, leveraging Google's apps like Maps, Lens, and Search. So the AI model behind Astra is described as multilingual, allowing users to interact in their preferred language with the AI providing vocal and written responses. So Google has also integrated a memory feature into Astra, enabling it to recall key details from previous interactions similar to the saved info feature in Gemini. And, yeah, we saw project Astra initially, released or teased, at IO, Google's IO conference many, many months ago. So still not out, but we saw some new features, pretty pretty impressive. Alright. Last but not least, I promise this is it y'all. So Google has unveiled deep research.

Jordan Wilson [00:07:46]:
This is huge. I freaking love this. Alright. I'm not gonna go ahead and say it's like a a a Suno moment for me or a notebook LM, but my gosh, I am loving deep research so far. So, Google has just released deep research. It's its latest AI advancement, and it promises to change how users conduct online research by automating the process. So it was just announced, and it's part of the Gemini advanced suite. So you do have to be on that $20 a month.

Jordan Wilson [00:08:14]:
Think of it like this. Here's what it is, and I'm gonna skip all my other bullet points. You know how perplexity can, you know, go through and, look at 15 to 20 websites? I love perplexity. I've loved it all along. This is gonna kill it. If perplexity doesn't pivot, I'm I'm dead serious. So, love perplexity, use it every day. I used Google's deep research last night to help me prepare for this show.

Jordan Wilson [00:08:42]:
My mind was blown. Okay? It visited a 169 websites for me. Let me repeat that. It I mean, it was a very long prompt. Right? And I gave it a lot of instruction, but it visited a 169 websites for me. It did all the browsing. Right? So, you know, OpenAI has their, you know, chat gbt search, perplexity, perplexity, fantastic. This could be a literal not I'm I'm not gonna use the word game changing.

Jordan Wilson [00:09:16]:
I'm gonna say the word human changing. I don't know, unless it's a quick one off thing. If I'm gonna be doing any other web browsing, I'm probably just gonna be using this new deep research. It is so, so good. So make sure you go check that out. But, again, you do have to have the Gemini advanced plan at $20 a month. I don't know why. I don't have it on my workspace account, which I pay for, but I have it on my, personal Gmail.

Jordan Wilson [00:09:43]:
I pay for the advanced version there as well. So it may not be rolled out if you are using, Google Workspace in your organization or your business account, but you might have to pay for it on your personal Gmail, but it is there. Alright. Oh, gosh. I told y'all. Hey. I warned you guys ahead of time. This was gonna turn into a dual episode of OpenAI and Google News, but alright.

Jordan Wilson [00:10:03]:
Let's get to this stuff you actually came for, the 12 new features. So what the what what is going on? Well, I think this is also a good example of go to market strategy. Right? Because no one was really talking about Google, right, until yesterday, and they went nutty. But we've all been talking about the 12 days of OpenAI. I love, what and and maybe this is just the start up mentality, but, OpenAI, I love what they've done with this. You know, they have a live stream every day at noon CST, 10 AM Pacific. It's good. It's entertaining.

Jordan Wilson [00:10:36]:
It's fast, and it's detailed. So, every single day, from, it started last week until I believe the last day is, December 20th. So next Friday, they're, releasing or announcing one new feature or one big update. So we are about halfway. Today marks the halfway point, so we're gonna be going over the first five. And I'm not just gonna be rattling off like, oh, here's what's new, but I'm gonna tell you what it all actually means. Alright. So day 1, here we go.

Jordan Wilson [00:11:03]:
And y'all, if you do have, questions as we go along, please please get them in. Alright. So day 1, we actually kinda got 2 announcements. They could've, if I'm being honest, they could've made this into 2 separate announcements, but that's fine. So on day 1, we got both Chat GPT Pro, which is a new tier, the elite cool kids club or country club of Chat GPT at $200 a month. So we got a new chat g p t pro. We got the, full o one model. Okay.

Jordan Wilson [00:11:34]:
So previously, this, OpenAI's new kind of reasoning model, which is completely different than their g p t models. We we just had a preview. We had the preview and the mini. So now we have mini, o one full, and then you have o one pro, but you only get that o one pro, if you are on the $200 a month, pro plan. So, there's some other features of that pro plan, when we talk about sort of which we're gonna be getting to here in a second. So, that's what was announced on day 1. So here's what these new modes and offers do. Right? So, the new o one updates.

Jordan Wilson [00:12:09]:
So, according to OpenAI, it offers a 34% reduced error rate versus previous models. It delivers deeper analysis and a PHD level reasoning for complex research, math, and coding, and it does allocate what they said extra compute time for the o one pro. So we don't even know is o one pro just a a like, a is it a completely different model? Was it trained differently? There's not a ton of information out yet. Essentially, what OpenAI said is, yo, we just give us some extra juice. We give it extra compute. Right? So that makes me think it's like, okay. Is the o one model being, like, somewhat throttled, or could the o one, the full o one model do what o one pro, is capable of? I don't think so. It it does just seem like, you know, there's just extra brains in this o one, pro model.

Jordan Wilson [00:12:55]:
So if you haven't used o one, I think it's limited use cases. Right? But if if you're in data, if if if you do anything around data, if you do anything around business intelligence, if you do anything around software development, if you do anything around, research, if you do anything in in the academic fields, I think o one is a great model. So, again, I'm not gonna go too deeply, but o one is completely different than, traditional large language models, what many of us use. So it does use this kind of, chain of thought or reasoning. So it thinks it thinks like a human. Right? And sometimes it will think for multiple minutes before it gives you a response. So it's not this, quick topical response that you're used to with the GPT series of models. It really thinks and it reasons, and it is really good.

Jordan Wilson [00:13:39]:
So what's the availability? Well, now. No wait list. It's available now. If you do want that o one pro, you do have to be on that chat g p t, pro plan, which is $200 a month. So what makes this unique? Well, there's a couple of things here because it's technically, like, 2 or 3 different things. Right? The chat gbt pro plan, well, what makes that unique is, well, it's the most expensive, kind of consumer grade subscription there is out there. Right? Yeah. We had, Devin's, you know, or Cognition's Devin that was just announced this week that we covered in our newsletter for $500 a month, but that's not for everyone.

Jordan Wilson [00:14:17]:
So this is, I would say, the first time that we've seen, an AI or a, large language model, related plan that is really premium. And I've been saying this for more than a year. I've said all along, there's going to be, AI plans that are 100 or 1,000 of dollars a month, as we get more and more capable models, right, as we, you know, maybe, officially or unofficially crossover from the HEI tier working toward the ASI tier. Right? Even as compute becomes too cheap to meter, we're still gonna have plans. And I do see this eventually, from, you know, maybe Claude, maybe Microsoft, maybe, Google. Right? But I I I don't think it's gonna be wild to think about that there's gonna be plans that cost 100 or 1,000 of dollars a month. Right? Especially as these models, become, you know, multi, agent environments. Think of reasoning models that have access to tools, that can perform actions on your behalf.

Jordan Wilson [00:15:17]:
Right? Who wouldn't pay 100 or 1,000 of dollars a month even if it's per user? But what makes the o one pro model unique is, well, it just gets extra juice. Right? And it's according to all the benchmarks. We already went over the benchmarks earlier. The benchmarks are very impressive. Very impressive at math, coding. You know, think of anything that a a a PhD would be, capable of. You know, that's what o one pro essentially is. It gives you, highly accurate PhD level, intelligence in your pocket.

Jordan Wilson [00:15:51]:
Wild. So how can you use it? Right? I told you I wasn't just gonna rattle off a bunch of stats at you today, although we're gonna be doing that. But, well, it can solve advanced scientific and mathematical problems with research grade accuracy. It can tackle intricate coding challenges, and it can execute tasks demanding careful, consistent reasoning. Alright. So, give it a try. Right? People are like, oh, should I do the $200, pro plan? Well, we're gonna talk about Sora, and maybe, the answer to that depends on if you're gonna be using sorta a lot. And I do think that we're gonna see more differentiation.

Jordan Wilson [00:16:25]:
I said this last week, the week before, or or sorry. I saw the I said this last week once the pro plan was announced. I'm like, there's gonna be more features that OpenAI announces during this 12 days that are gonna be, extremely, or or sorry, that the paid plan is gonna be limited. The $20 a month plan, it's gonna be, limited for a lot of these new features that we saw. Saw that immediately with Soera. You don't get access to everything on the $20 a month plan. So, the $200 a month plan, there's gonna be a lot. But, you know, even on the $20 a month plan, you can get your, get your feet wet with o one.

Jordan Wilson [00:16:58]:
I still use o one every day even though I'm not a a researcher. I'm not using it for, you know, PhD level mathematics or coding challenges. Right? I'm dumping a lot of data on it, and I'm having it think for me. Right? It's weird now. So often, at least when it comes to business, I I I always have multiple large language models do my first round of thinking for me when it comes to a a deep task. Even even something like planning today's podcast. Right? I go to reasoning models first. Right? It's like, oh, how should I open it? How should I plan it? How should I structure it? Right? Here's all my data, all my notes.

Jordan Wilson [00:17:34]:
Right? Because I'm taking notes. Put it all in a in in a good format for me. Right? So it's it's not just if you're a a mathematician. Alright? So day 2, Sora. Alright. We covered, did did did a little review of Sora on our YouTube channel, and we shared that in, in our newsletter as well. But, hey, livestream audience, let me know. You guys interested in Sora? It's it's weird.

Jordan Wilson [00:18:06]:
I've you know, in some previous lives, I did a lot of video shooting, a lot of video editing. You know, I don't really do any of that anymore, but it's really good. It's really good. And, you know, and I keep thinking in my head, who is Sora actually for? Right? And I think very early on, creatives. Right? So, but I think, eventually, everyday people are gonna be using it. Right? As content becomes easier to create with AI, I think consumers' expectations for highly personalized content is going to increase. Right? So I I would say, let's just say let's just say, for for easy, use. Right? Fortune 500 companies.

Jordan Wilson [00:18:52]:
500 companies. Let's say only 10 of them really leaned heavily into, creative video production, traditionally. Alright? Now I think it's gonna be the other way around. It's gonna be 90%. Right? Because consumer expectations are going to increase because we're gonna start getting highly personalized, very high quality video content that speaks to us. Right? So I think, what Soera is starting to bring, is is gonna change it. So, yeah, Soera, if you don't know, it is OpenAI's text to video model. This is one of those ones.

Jordan Wilson [00:19:25]:
Talk about wait list. Right? I I sometimes call Google out. You gotta call gotta call OpenAI out on this one. Right? And then I know there's reasons, you know, they they preview this in February. So this was a 10 month, wait list. And I said I said in February I said in February, this isn't gonna come out until after the US election. It didn't. Right? So there are certain things when it comes to safety, you know, even when it comes to, following laws.

Jordan Wilson [00:19:50]:
Right? So as an example, Miriam here is saying no sorta here in the UK. Yeah. CEO Sam Altman said it's gonna take a while. Right? There's certain laws, in certain countries that might prohibit things and it's for safety reasons as well. So I get it. Super long wait. But let's talk about what sort of does. So it transforms detailed tax prompts into HD videos up to a minute long.

Jordan Wilson [00:20:13]:
If you do have that $200, a month plan, the pro plan. It creates, it creates, extends, and refines scenes with multiple characters and precise motions, and it maintains high visual fidelity, and prompt adherence. So, is it perfect? Absolutely not. Right? No. Yeah. Try try something that requires real world physics. Not good. Right? Try to run a sore prompt with, something with gymnastics.

Jordan Wilson [00:20:41]:
Right? Something with a human, you you know, or a group of people mountain climbing. Right? The physics still aren't there. So you you do have to understand what Sora's strong suits are capabilities. If you're gonna start with tax, if you're gonna start with an image. But if you know what you're doing, and you might have to run 4 different variations to get one that works. But for the most part, if you know what you're doing, you can create video that 5 years ago would have taken 1,000 of dollars, multiple humans. You can create something decent in a couple of hours, right, by stitching some of these short clips together. You could create a promo video for your business.

Jordan Wilson [00:21:19]:
You could create, an advertising, campaign. You can create so many things that, again, would have taken a big budget and a lot of humans. You can do that with a $200 a month subscription now. Right? I don't think you could do it with a $20 a month because your credits get burned quickly, and you do sometimes have to run multiple generations to get something usable because you get something different every time. But it can it can, again, I'm not gonna say replace, but it can really do what 1,000 of dollars and multiple humans used to do 5 to 10 years ago. Extremely impressive. Well, the availability, it's availability to all ChatGPT Plus and Pro users, in future tailored pricing is planned. So, yeah, we might see a a a credit system because, you know, I saw online even people that have the chat gbt, you know, pro plan, $200 a month.

Jordan Wilson [00:22:13]:
It is a credit based system, and you essentially have unlimited, generations of lower quality. But if you want HD quality, you're gonna run out. You could even run out pretty quickly on that $200 a month plan. Right? Where I think, you know, other video generators where they, where they might excel is allowing you a bigger plate. Right? With a runway, you can eat more. Right? With the Luma Labs, you can eat more. Right now, I think Asura is higher quality than some of those. People might, disagree with me on that.

Jordan Wilson [00:22:47]:
If you look at the ceiling. Right, if you're not looking at the average because, tools like Runway, like Luma, like Cling, all of these other AI video generators, they've gotten, you know, millions of data points. So it doesn't have that yet, but they will soon. Again, this is the worst sort of will ever be. But I think Sorra has a higher ceiling. The average, output right now, Sorra might not be the leader, because it's a couple days old, right, in terms of the general public. Right? OpenAI servers have been crashing. Like, I I saw I I read online December has been, OpenAI's, most or I I guess their hardest hit month in terms of uptime.

Jordan Wilson [00:23:26]:
Right? Everyone's rushing and crushing the server. So, it will hopefully get a little bit better. Yeah. A couple people asking about VPNs. Right? So, yeah, if you're in Europe, certain certain countries in the EU, you don't have access. If you use a VPN, you could try. Right? But they explicitly said they are going to ban accounts that are doing that. So I wouldn't necessarily recommend it.

Jordan Wilson [00:23:54]:
Alright. So, so what makes, Sorra unique? Well, there's a lot of features that I like, that are super easy. So I'm gonna talk about a couple here. So one style presets. Not super unique, but I think it's unique in the way that it's implemented. So you can very quickly you don't have to be a a prompt, prompt engineering expert. You don't have to be an AI video, aficionado, to be able to really get the most out of with some of these presets, but I love the storyboard feature. I think it it's not super mature yet, but I think it's gonna get better.

Jordan Wilson [00:24:29]:
It did play around with it a little bit, but this is essentially a way that you can make a coherent, story. Right? Because that's been one of the biggest challenges with with AI videos so far, is is having, being able to piece together multiple of these clips. Right? Because if you try to let it create something long, it may go off the rails. So, generally, your best generations using AI video tools are running shorter ones that you have more control over and probably starting with an image versus text. Right? So what a lot of people have done is, you know, there's now AI image tools that allow you to create consistent characters. So then they just do that and then create, you know, line those up and create multiple videos. But with the storyboard feature, it does that. Right? So it's like you can literally run multiple text prompts, you know, one after another after another after another, and Soar does its best to make them all work together.

Jordan Wilson [00:25:19]:
Super impressive feature. So how can you use it? Well, we already talked about this, but you can use sort of quickly produce concept visuals and unique animations. You can generate educational or marketing videos with minimal effort. You can streamline content creation, saving a ton of time and resources. Alright. Alright. Let's keep it going. Let's keep it going.

Jordan Wilson [00:25:41]:
Liz Liz said, Sora broke my heart. They're on GPT teams, and you didn't get access to it, but Plus did. Yeah. Teams, teams and enterprise, clients aren't always gonna be able to get access to all of these things. Right? I have multiple free accounts. I have, ChatGPT Plus accounts. I have ChatGPT Teams accounts. I I have enterprise.

Jordan Wilson [00:26:05]:
Right? Because clients hire us to help, help them help help their teams learn chat gbt. So, yeah, some of these are are if you want access immediately, you might just have to be on the chat gbt plus plan, or on now on the pro plan. Alright. Day 3. Alright. We're gonna go a little faster. This one's dragging on. Google Google had to screw up the timing and just ship like crazy.

Jordan Wilson [00:26:28]:
Alright. So day 3 was reinforcement fine tuning research program. Little technical, but super cool. Alright. So here's what was announced. Well, it was an expansion of the, r f t r p, or r f. I I I don't even know if if if we're gonna, abbreviate that one. So the reinforcement fine tuning research program.

Jordan Wilson [00:26:48]:
It was an expansion of the program enabling reinforcement based custom AI model development. I know this sounds dorky. It's not as dorky as it seems. So here's what it does. It employs reward driven loops to refine model reasoning and accuracy. It's you can create specialized models in a narrow domain such as medical or legal with minimal data. That is the key here. We're talking about dozens of examples, not 100 of 1,000, right, when you would talk about traditionally, fine tuning models, right, with, take a long time.

Jordan Wilson [00:27:22]:
So now with this kind of new program, we're talking about dozens of examples. Alright. So it essentially creates expert models with significantly improved task performance. Right now, it is in alpha, so it's not even in beta phase yet. So, you you know, I wouldn't see, plan for general availability for this anytime soon. Although, OpenAI did say that a the release for this is planned for early, 2025. So here's what makes it unique. It uses a grader that you can reward desired outputs, boosting correctness, and reasoning quality, and it requires, like I said, way fewer examples.

Jordan Wilson [00:28:01]:
So think of, this is not, sometimes y'all like this shows for everyday people. Sometimes I sacrifice 1% of accuracy so I can get 99% more clarity. Okay? Think of it like this. 2 years ago, if you wanted to fine tune a model, right, so that means make this big model specific for your domain, It would cost a lot of money, sometimes 1,000,000 of dollars. It would take a lot of engineers, and it would take sometimes 100 of thousands of your own data points to essentially train the model. Right? So think of it as you're using a car's engine. Right? And then you're building, you you know, the tires and, the I don't know anything about cars. I shouldn't use that as an example.

Jordan Wilson [00:28:49]:
Right? But you you know, you're building all these other pieces and it takes time. Now it's not like that anymore. Right? So what this could lead to is customizing your own car, but with a couple of clicks. Right? And not having to have a team of people building the rest of the car just starting with the engine. So, you you know, how we can use and build these models is going to really change. And, being able to get a fine essentially a fine tune model with only a couple dozen examples, right, where you have to train the model. Hey. When someone asked about this, this is a right answer.

Jordan Wilson [00:29:25]:
This is a wrong answer, and this is why. Right? You essentially easily can classify those things. It makes it simple. Right? It makes it to to where nontechnical people and smaller businesses, startups can start using this once it's available. Whereas, normally, in order to fine tune models, you had to be an enterprise company with a bunch of, you know, AI engineers essentially. So here's how you can how you can use it. You can produce high accuracy models and focus fields. You can improve outcomes on complex domain specific task and achieve strong performance with fewer examples.

Jordan Wilson [00:29:59]:
Alright. I see I I see a couple of questions y'all. Keep keep them coming. I'm gonna tackle any of the questions that you have, at the end as I'm scrolling through here. So let's talk about day 4. Alright. So day 4, Canvas updates. So here's what was announced.

Jordan Wilson [00:30:17]:
Canvas was previously only for chat g p t plus users. Now it's available for all. So if you are using the free version of chat gpt, which I would not recommend, just pay the $20 a month, y'all. It's it's the easiest, like, you know, think of your coffee. I'm sipping a coffee right now. Hey. Shout out Nespresso. I love Nespresso more than anything else.

Jordan Wilson [00:30:38]:
They should sponsor the show because I'm always sipping, sipping on their coffee in the morning. But think, if you go get a coffee at Starbucks or your local coffee shop, let's say you do it twice a week, 3 times a week, well, there's your $20 a month chat gbt plus or yeah. Chat gbt plus subscription plan. You know, people, if if they're using the free version, they just give you enough to get a taste, but it's not enough to actually get work done if that's what you wanna do. Anyways, so, the big update with Canvas, well, it's available to everyone. There's some other bells and whistles. Essentially, Canvas was previously its own mode, but now it is not its own mode. So it is used in the tools section.

Jordan Wilson [00:31:17]:
So there's now a new icon when you are using ChatGPT, and, essentially, it is a tool that you can use inside any mode now aside from the o one models, in advanced voice mode. But, essentially, it's not its own mode anymore. So it's a tool. You click on it to use it. The other big update here is it can also run Python, and it does now work with custom GPTs. Alright. So as I go over what it does, hopefully, that'll explain a little bit more. So Canvas serves as a side panel within Chat GPT for collaborative writing and coding projects.

Jordan Wilson [00:31:51]:
You're not collaborating with Teams, right, like in Microsoft Pages. You're collaborating with Chat GPT. So, normally, when you are conversing with Chat GPT, right, you give it a prompt, it gives you a long answer. If you wanna change something, you gotta kinda reprompt, but Canvas changes it. So think of it as a split pane. So on the left side, you're talking to ChatGPT. On the right side, it's like an editable document. That's a hard word to say at 7 AM in the morning.

Jordan Wilson [00:32:17]:
Editable. Editable. Alright. You can edit the document live. Right? And you can also work with ChatGPT live. Right? So think of in a document, you can go you can type something yourself, tag ChatGPT in without having to constantly re prompt, scroll up, you know. It's it's it's just a much better experience. Right? I actually like that they called it Canvas because that's what it is.

Jordan Wilson [00:32:42]:
It's a Canvas and you can stay in that Canvas and work on it and chat gbt. Right? There's all these nice little, you know, one touch buttons, you know, to polish the writing, change the writing style, do code debugging. Right? But you can stay there. So it's just much easier to navigate with this kind of dual pane. Right? So you can, essentially, you know, kind of quote unquote command chat gbt on the left pane, and then on the right pane, you can see it work. You can go in there, work with it, change some things manually. It's it's a great mode. So it does allow direct editing of text and code, support shortcuts, and keep version history.

Jordan Wilson [00:33:14]:
Also, now the big thing is you can run Python code. So this is a shot at artifacts. Right? I do think in Anthropic's Claude, the artifacts feature was its best its best feature. Right? That's 80% of what I use Claude for, is for artifacts. Claude still has, and artifacts still has, way more advantages because essentially, it can run any code. Artifacts can or much, you know, I don't have a complete list. But you can run, you know, you can create an HTML website. You can create a business dashboard.

Jordan Wilson [00:33:47]:
Right? You can do all these things. So right now, the new Canvas updates, you can just run Python. Right? So you can't just render out any type of code, but Python's a great start. Right? And this is something at least if if if you're, an engineer, software developer, coder, business intelligence, etcetera, you're using Python. Python is an is an extremely flexible programming language. You can create graphs with Python. You can create games with pie right? You create so many things with Python with that programming language, but, Cloud Artifacts still has an advantage if you're just looking at a large language model's ability to render something in the window. So it's available now.

Jordan Wilson [00:34:25]:
Right? So, yes. Canvas available now even if you are a free user, and it's available inside of GPT. So you do have to have a paid account to create GPTs, but now you have that kind of canvas feature or canvas ability inside of your GPT, which I think is really nice. So what makes it unique? Well, it, like I said, provides a visual side by side interface for working directly within ChatGPT. ChatGPT also gains better context awareness, offering targeted inline feedback and suggestions. You know what? I actually haven't done, context text, context testing, within, ChatGPT, which I probably should do. Right? So I'm assuming that it's not going to eat up as many tokens, if all it's doing so let's say, you know, you you asked chat gpt to spit out a 1,000 words, let's just say, or a 1,000 tokens, right, each time, and you're just slightly reiterating. Eventually, you're gonna hit that 32,000 token limit, in chat gpt on the front end.

Jordan Wilson [00:35:29]:
Right? If you're using ChatGPT on the back end, it's a 128,000 tokens. That's its memory before it starts forgetting things. So I do need to do some testing because I'm assuming, that the canvas using the canvas feature, is technically gonna eat up a lot fewer tokens. So even if if even if you don't think that you're gonna need that, direct editing, feature, I don't necessarily see a downside, to using Canvas most of the time for for for most everyday use. Also, what's great about, Canvas is you can highlight specific sections for focused assistance, and also you can there's those nice little shortcuts on the side. Right? Where if you're working with code, it can debug, suggest improvements. You can change the writing tone with one click. You can make it longer or shorter.

Jordan Wilson [00:36:16]:
Right? Very cool. Just easy button, you know, like huge one click, that can change the output. So, how can you use it? Well, you can collaborate interactively with ChatGPT on writing or coding tasks. You can run and refine Python code right in the workspace for immediate validation, and you can request and receive targeted edits, suggestions, and improvements on your work without even having to talk to Chat GPT. Right? Sometimes you're like, oh, okay. How do I, you know, have Chat GPT, you know, change it this way. Right? And you're thinking, and you gotta type it out, and you're like, ah, that's not right. Right? And you're like wondering, how can I give Chat GPT feedback? Well, in the Canvas mode, which is I don't think anyone else has that level of one click flexibility.

Jordan Wilson [00:37:02]:
Right? There's there's other large language models where you can, you know, change the tone, make it longer or shorter with one click. But Chat GPT in the canvas mode, really brings a lot of easy one click modifications, to that toolbar. So like I said, if you're not using canvas mode, it's pretty much my default mode even though it's not its own mode. Now it's a tool, but there's I see very few instances where you shouldn't be using it as your default. Alright. And now we have our last day. Here we go. We got to the end.

Jordan Wilson [00:37:31]:
Day 5, y'all. So it's the Apple Chat gpt partnership. Is this new? No. Nothing was new here. Nothing was new except it's out. Alright? So, you know, I think all these other things, there's been a certain level of, surprise or intrigue. So, you know, day 5, not so much. It was just essentially the Apple and Chat GPT partnership is live.

Jordan Wilson [00:37:54]:
Alright? As Apple, updated their iOS and their Mac operating system. So here's what it does. We've covered it very, very, very in-depth here on the show. So it embeds, ChatGPT into Siri, and also Apple's writing tools. It uses essentially ChatGPT. So there's settings that you can go into, on your iPhone and enable this, and or, ChatGPT will ask you or sorry, Siri will ask you. So if you're asking Siri, something that is a little too complex, it will say essentially, do you want to use ChatGPT for this? Or you can just, kind of override that and make that your default setting, within Apple's settings. So, it does, embed chat gbt into Siri and Apple's writing tools for advanced reasoning and content generation, can analyze images, in documents natively with no extra accounts needed.

Jordan Wilson [00:38:45]:
Also, this new integration that's live enhances Siri's intelligence beyond standard assistant functions. Are we finally gonna have a smart AI assistant? Well, we'll see. I think my devices, I didn't even have time to update them yet. Alright. So, here's the availability. It's availability. It's it's available now. Right? So, you do just have to update to the latest iOS and the macOS updates.

Jordan Wilson [00:39:07]:
A lot of it is device dependent though. So, you do as an example, if you're using it on your Mac, you have to have an m chip. So an m 1, m 2, m 3 chip. So it's not gonna work on super old, Macs. The same thing on the phone, you do, to use the kind of the camera or the vision feature. You do have to have, the newest iPhone. So that's not gonna work with all of them. Alright? Because, you know, for some of these things that happen, you know, on edge AI.

Jordan Wilson [00:39:35]:
Right? It's happening on your device. Some of these things happen in the cloud, so you don't need the most powerful, device. But some of these things you do need, the most powerful, processors. Yeah. Monica here is saying finally. Yeah. I agree. I'm I'm excited to, test out that new feature, and and to report more on it.

Jordan Wilson [00:39:54]:
Because to tell you the truth, I haven't I haven't been able to research that one as much. That is the one, the last announcement that just came out. So, what makes this unique? Well, one thing that Apple does a really good job on is privacy. Right? Yes. They are last to the game. I think most people, know, by the time they hear Apple Intelligence, at least if you're like me, I kinda roll my eyes. You you know, I don't know. To be honest, I I kinda hate the the the hard stance on branding that Apple is taking toward artificial intelligence, by calling it Apple Intelligence.

Jordan Wilson [00:40:26]:
I think that's a little, little cheesy, especially coming from the company that is multiple years behind its biggest competitor in Microsoft. Right? So to try to rebrand artificial intelligence as Apple Intelligence when most people still aren't gonna get these full features until 2025, and we've already had, Microsoft Copilot for multiple years now. Kinda cheesy Apple. I don't know. Alright. So here's how you can use, this new chat gbt Apple intelligence integration. So you can draft documents, emails, and, analysis without leaving Apple's ecosystem. You can summarize PDFs and interpret images right on your device, and you can enjoy faster, more integrated AI assistance for everyday tasks.

Jordan Wilson [00:41:12]:
I'm not gonna use it if I'm being honest. I'm not gonna use Apple Intelligence, at least 90% of them. I mean, we'll see, if I end up using Siri. I I just like using advanced voice mode on chat gbt. Also, Google Live, their voice mode is probably the leader right now because it is the only voice mode that is, neural voice. Right? I'll say that because obviously Apple and and Siri or or sorry, Siri and Alexa are, connected to the Internet, but they're dumb assistants. Right? So the only true, like, AI assistant that is connected in real time to the Internet right now is Google Gemini's live, but hope and now Siri, technically. Right? So I haven't tried it out now, but, hopefully, we'll see some updates from the advanced voice mode.

Jordan Wilson [00:41:57]:
I love using it, but the downside of Chat GpT's advanced voice mode is it's not connected to the Internet. So you're not getting the most up to date real time information. Alright. So that's how you can use this new Apple, kind of, thing here. So let me just quickly, I saw a couple of questions. Sorry if I missed it. I'm I'm actually on my laptop today. So normally normally, I have, you know, 2 computers and I can see a little bit more.

Jordan Wilson [00:42:21]:
So I I'm I'm gonna try to get to a a couple of these questions here, and then I'm gonna tell you what I think is next for OpenAI's kind of 12 days. So Frank is asking, do I have to pay per user for pro if we are in the same organization? Frank, I believe, yes. Although, I don't believe, I don't believe as of yesterday, at least. I was talking with one of my contacts at OpenAI. I don't believe that, pro is yet available, for Teams or enterprise accounts. I'm sure it will be any day now. But as far as I know, right now right? And I have a Teams account. I have an enterprise account, but I I don't have billing or admin access to that.

Jordan Wilson [00:43:03]:
But on my Teams account, I can't upgrade to pro. I can only do that with my kind of quote unquote normal accounts. So, you you know, a free account, you can upgrade to pro. A chat gbt, plus $20 a month, you can upgrade to that $200, account. Frank also asking where did 01 go? I had it 7 days ago. For us it's it's still there. So, yeah, if if you, if you have a paid $20 a month account, o one is still there. Philip joining from YouTube says Gemini Pro and Ultra.

Jordan Wilson [00:43:34]:
So can we say Gemini is the biggest competitor against OpenAI? Yeah. They are. I'm gonna save this rant for another day. Philip. I think I think wow. I mean, if I'm being honest, I think until yesterday, OpenAI was in its own tier. Right? It was tier 1 OpenAI, tier 2, you know, 2 a Google, you know, 2 b Microsoft even though Microsoft just essentially no. I gotta change that.

Jordan Wilson [00:44:09]:
I would actually say previously, I would say one a OpenAI, 1 b, Microsoft Copilot, then I would say 2, 2 a was probably, Google Gemini, and then, you know, 2 b was probably a claw to Anthropic. But now I think Google is back up in that tier 1, with what we just saw yesterday. I mean, my gosh. I mean, the problem, they have a marketing problem. Google has a go to market problem. OpenAI does not. Right? OpenAI brings again, yeah, sometimes you get these long wait list, but at least the products get released. Google, not so much.

Jordan Wilson [00:44:42]:
And sometimes they're just releasing it on the developer end. Right? Which people don't understand. People do not understand this. 5 years ago, it was CTOs, CSOs. It was the highly technical people that were making these decisions. So I think Google especially has this all wrong. They have their go to market all wrong. They're only appealing, I think, or or relying too much on developers.

Jordan Wilson [00:45:09]:
Because aside from, you know, finally, the front end of Google Gemini just got some much needed updates, in the last 24 hours with the new, 2 o Flash and also the new 15 Deep Research. Amazing. But for the most part, Google the the the the front end of Google Gemini has just been abandoned. Right? But that's what people use. I literally literally and this is not an exaggeration. I people reach out to me all the time, and they're saying, Jordan, listen to your show, blah blah blah. You went off on some crazy rant, but our entire executive team looked it up and you were right. Right? But these executives that are trying to make decisions on AI, they're going on the front end.

Jordan Wilson [00:45:48]:
They're going chatgbt.com, gemini.google.com, claw.ai, Copi. Right? That's what they're doing. They're not going to Google's AI Studio or or or Vertex anymore because now the the decision making, ability, it is CEOs. It's CMOs. Right? It's people that are not necessarily technical. They're not the ones necessarily going on the developer platform to test things out. I think Google has tremendously missed their go to market. But, yeah, right now, they're the biggest competitor.

Jordan Wilson [00:46:22]:
Sorry, Philip. I went on a rant. Alright. Jay, how long did it take to give you a response after search the 160 sites? Great, Jay. So, yeah, it searched, the Google deep research, searched a 169 sites. I literally like, I made a face. If you're on the livestream, I was like this. I was up here by myself.

Jordan Wilson [00:46:44]:
I don't know. 10:11 PM last night doing some last minute prep for today's show, and I I I was legit flabbergasted. It took probably I should have timed it. I'll probably do a video review. It probably took about 2 to 3 minutes. I'm guessing it might take longer as more people use it. Right? It might be kinda like, you know, bandwidth. Right? So if you're using it during busy times, it might take longer.

Jordan Wilson [00:47:07]:
But it took about 2 or 3 minutes. But you know what? I'm fine with that. Because as I've been using perplexity, I've been using perplexity since it came out. I personally feel in in in my team's experience, perplexity is getting more things wrong than they used to. It's new kind of, quote, unquote, reasoning, not that good. I don't think this new deep research, I was blown away. And, yeah, Jay, I'm pretty much ready. I'm pretty much ready to drop perplexity, not just yet.

Jordan Wilson [00:47:37]:
Right? There's a lot of things that I had, you know, baked into our workflow, that I'm not yet ready to drop perplexity because of this new, deep research. Have to do a lot more testing. But, I mean, instantly, 80% of my use cases for perplexity went out the window. The very first time that I use deep research, from from Google. Amazing. And that's not really happened. Right? Very rarely do I use a tool once. Right? Because I I test I stress test these things, try to break them, you know, reverse engineer what's good and bad.

Jordan Wilson [00:48:11]:
Right? Very, very hardly ever do I use something once and say this is gonna replace a big part of of my team's workflow. That happened. That happens. Alright, y'all. I don't see any more questions. I hope this was helpful. I hope it was helpful, to to get a nice, overview of everything that's been going on, but here's what I think is next. Okay? There's still 7 days.

Jordan Wilson [00:48:41]:
So what is OpenAI going to release? Well, I think we're gonna see projects, and and and some of these, you know, there's some random rumors on the Internet, in in screenshots, and some of these things are just my hunch from covering this every single day. I think we're gonna see projects, which is essentially a a folder system inside Chat GPT to help you stay better organized, but also to upload files. So something between, like, spaces in perplexity and projects in Claude, but essentially one spot where you can upload custom instructions, different chats, and different files. I think we're gonna see that. I think we're gonna see some advanced voice mode updates. Hopefully, the ability to use tools in the Internet, within advanced voice mode. Right? That's what I want. I wanna be able to have this neural, voice that I can talk to like a Siri, like an Alexa.

Jordan Wilson [00:49:28]:
Right? That's actually smart, that is powered by a large language model, but I can upload my data to, and it can also look up real time information on the Internet. Because right now, advanced voice mode, you cannot upload your own data. It doesn't have access to any of the other tools of chat gbt. So I think and I hope that we're gonna see something like that. I think we are gonna see a a gbt 4.5, maybe not released, but a wait list. I'm not sure. But I think we're at least going to see the next GPT model since we've seen a lot of, movement in the o one series. And and I think we're gonna get a preview of the rumored operator agent from OpenAI, especially, right, what we just saw from Google.

Jordan Wilson [00:50:10]:
We've been seeing reports for many months that OpenAI has been working, on this new, agentic system, and that's next on their step. Right? They have this 5 step plan to AGI. Right? And we're at reasoners, and it's so it's it's only, it's only logical that the next step toward that is agents. Right? We've seen it from Microsoft Copilot. We've seen it from Salesforce. Right? We've seen it, now from Google. Right? Even though it's probably gonna take many, many months for most of us to get our hands on this, but I do think that's next. So that's what I think.

Jordan Wilson [00:50:40]:
Let me know. Actually, right now, if you're still listening on the livestream, get your best guess in for what's next. I'm gonna grab some of my favorites, and put them in today's newsletter. Alright. So, let me know what you think is next. Alright. And, also, if you haven't already, please go to your everyday ai.com. Sign up for the free daily newsletter.

Jordan Wilson [00:50:58]:
If this was helpful, please repost this. I know this was a longer show. It ended up being half Google, half OpenAI. But, hey. In 50 minutes or in 25 minutes, if you listen to me on 2 x, which I definitely would, you just got tons of hours worth of updates, and now you are the smartest person in AI at your company. Alright. So thank you for tuning in. Again, please go to your everydayai.com.

Jordan Wilson [00:51:23]:
Sign up for the free daily newsletter. Yeah. We recap today's episode in every episode. So when we have super smart guests, we recap the main insights in our newsletter as well as keeping you up to date on everything else that's happening in the world of AI. Thank you for tuning in. We hope to see you back tomorrow and every day for more everyday AI. Thanks, y'all.