← AI For Humans: Weekly AI News, Tools & Trends
AI For Humans: Weekly AI News, Tools & Trends
Claude Sonnet 5 Is Here & Fable 5's Returning. For Now.
00:24:20
Loading summary
Transcript64 lines
- [00:00]
A
There's a new Clod Sonnet 5. And maybe we'll get the return of Fable 5 as well. But.
- [00:06]
B
Ah, yeah, Big but. And my anaconda don't want none. OpenAI's new model, GPT 5.6 Sol, isn't going to be available for who knows how long. Because now, Gavin, we live in the weird realm of jailed models.
- [00:22]
A
This is me inside jail, Kevin. We would break these models out if we could right now, but in the meantime, we're going to be diving into some of our favorite AI video workflows. These are things that Kevin and I are actually doing today that you can do as well.
- [00:34]
B
And there's an amazing new blender to seed Dance workflow, which lets you control the camera with incredible accuracy. You're not prompting videos, you're directing them. You can turn basic blocks into gunslingers, or you could try to turn these blocks into something wholesome.
- [00:50]
A
Censor that for now, will. We have so much more AI news. We are humans. This is AI, and therefore this is
- [00:56]
B
AI for human and trapped stepmothers.
- [01:02]
A
Oh, no, no. Foreign. Welcome, everybody, to AI for Humans, your weekly guide to the wonderful world of AI. My name is Gavin.
- [01:12]
B
Priscilla.
- [01:13]
A
That's Kevin Prayer. And Kevin. Today we are getting rumors of a new Sonnet 5, and this has been rumored for a bit. It may drop today or tomorrow if you're listening to this. But the other thing about this, Kevin, I think that's important is that these rumors come out. It's like trickling out because people are jonesing right now. We are. I am jonesing. I think you were probably jonesing, too. Fable 5, still not available. Sonic 5. The story on this is it's going to be much cheaper than using opus. It's going to be more expensive than previous sonnets, I think. We don't know for sure yet, but overall, they're saying that it is performing very well. And I think it'll be a good model when it or if we are able to use it. Breaking news. Right after we recorded this, Clod Sonnet 5 did come out. The model looks very good. Benchmark. Boys are feeling better about this. It compares very nicely to GPT 5.5. You can take a look at some of these benchmarks. It is literally just dropped. It just showed up in my model picker. But you should have access to this now. I think one of the coolest things about Sonnet 5 is going to be this idea that, like, hey, whatever it was three months ago when 5.5 came out, that was the frontier. And here we are now. So I'm really excited to drop this model into some of the projects I've worked on, especially ones that deal with inference that need to get information back from the model itself. So the fishbowl, that thing I made a couple months ago, which is still up, you can go try it. Fishbowl show was running off of Claude Sonnet 4:6. This is going to make it much, much smarter. I'm very excited about that cost wise. I think this is going to be interesting to watch because if we are now getting the intelligence of GPT 5, 5, which again was the frontier, say three or four months ago. Three months ago. Now three months later, we are getting that intelligence for significantly cheaper. Again, this is going to be a brick in the wall of like the AI Ha. Where we are getting faster, quicker, better intelligence over time. But we are going to get into this right now about what it means that we don't have access to the cutting edge models right away. And whether or not we will get them anytime soon, I don't know. This is breaking into breaking news. I'm breaking into the breaking news from before that Sonnet 5 came out with brand new news. This is later in the day when we recorded all of this stuff has happened. What a busy day. Fable 5 is coming back tomorrow. At some point when you Listen to this, Fable 5 will be live again. We do not know a lot of details if it will change at all, if we will have to do any sort of like sign in if it's going to cost less or more. But Anthropic has been Approved to release Fable 5 back into the world. So get ready for this weekend. You can do some fun building. Uh, I think this is really interesting. Like obviously OpenAI is going through the same process too. I assume now that Fable 5 is being released, we are going to see OpenAI's 5.6 all come out sooner rather than later. This kind of changes a lot of the conversation that we are having in this show about this. But still the idea that the government is having some sort of control over these before they come out is a big deal. We'll talk a little bit more about what that means in a second here. But more importantly, rejoice, everybody in the world rejoice. Frontier AI is coming back tomorrow, at least for Mythos and Fable 5 and Anthropic, hopefully crossing my fingers that that means GPT 5.6 SOL, which we have not used yet, comes back soon as well. All right, I'm going to go back to the breaking News from before and then we'll see what happens. But here we are. Back to us.
- [04:48]
B
I haven't shaved. A because lazy and B because you would see the nail marks on my neck I have been scratching so fiendishly waiting to larger products, more and more ambitious projects until these models improve because
- [05:03]
A
we had the taste.
- [05:04]
B
And then it went. And so a couple things like we have some rumors that are being echoed pretty loudly so we, I think we can report them with some fair conviction. We are recording this on an early Tuesday your time. So by the time this hits, maybe there'll be updates with the sonnet of it all. The rumor isn't that, oh my God, it's so all powerful and all being and this is going to make you forget Fable. The reason people are even remotely excited for this is that it's going to perform supposedly at the level that Opus does today, but it's faster and cheaper. And why does that matter? Because everybody else, like I know that we have some casuals in our audience and hello, please try to stay with us here. Everybody else is talking about this GLM model, this other model that came out of China that punches at the weight of the foundational models that we have access to from anthropic and from OpenAI and while the government restricts access to things and we'll get to that, I think even more in a second while that happens here stateside, everybody seems to be jumping ship to adopt these Chinese models through OpenRouter or to run them locally on their machine. And they're cutting their inference costs. They're dramatic. They're cutting millions of dollars, some of these massive companies out of their AI spend. And if you're Anthropic or if you're OpenAI and you're sitting there going, well, what's the answer? If you can't release the strongest, most foundational something, you better release something that's fast and cheap to compete with the, you know, at least half as expensive model that everybody seems to be switching to.
- [06:32]
A
Kevin There is a pretty strong rumor and again, these are speculative rumors. We don't know for sure, but there is a strong rumor that there may be a login where you have to prove who you are in order to use Fable 5 KY KYC know your customers we talked about last time. Kevin I think the big story that came out on Friday, which is still percolating right now, is that OpenAI announced a new model, GPT 5.6 Sol. This is a brand new flagship model, but according to Sam Altman, they are now having to go through the same governmental process that we assume Fable 5 is going through after Fable released. This model, according to the benchmark boys, is very, very good. But we have a new Benchmark Boy song that we are going to make right now. It's more of a funeral dirge than it is a. An actual. So let's listen to that real fast. That's right. It's a sad benchmark Boys, because we don't have access to this. You don't have access to this. And who knows when we will get access to it. According to the benchmarks, though, Kevin, this model outperforms mythos 5 according to benchmarks, so we will see what this looks like when it comes out. I gotta imagine if you're inside of OpenAI, this is kind of a bummer moment. It's a bummer moment a little bit for me and maybe for you as well, because, like, you used to have this sense of like, hey, here's the new model. Go play with it. And now we're like, here's the new model. I guess it's good. I have no idea. We don't know what it can do other than the 20 companies. Only 20 companies have access to it right now, supposedly.
- [08:11]
B
Well, that's the thing, is that it's like, here's the new model. And by the way, like, they're all going to play with the carrot and you just keep getting the stick.
- [08:17]
A
Yes.
- [08:18]
B
So, like, we're not going to get into the inequality of AI access on this particular episode or the haves and have nots, which we have, you know, been concerned about for a long time. Two questions for you. One on the KYC of it all, the know your customer. Yes. What. What data are you willing to hand over? And maybe the answer is all of it, but, I mean, Sam Altman has an iris scanning orb. We know that. Like, what level will you go to and what level do you feel comfortable proving you are who you are to get access to foundational models?
- [08:53]
A
I mean, I guess the question is in the same way that, like, I want to be able to prove to the banks that I am who I am because I want access to my money. I think in some ways that's what this is going to become. I don't know if mirror scanning, maybe if it's easier and it's private and I can control it. But, like, I do think that this is going to get to that level at some point where people will start to consider the security of these things. And I honestly, like, I said on the show before, I don't have as big a problem as many people do with privacy issues. I also think, like, it really depends on what you're using these models for and what you want to do with them, in part. But there are a lot of people out there who just won't do this and maybe they will send more people to those Chinese models. And Kevin, we are going to talk later about. We said this before, but like Sea Dance and their new audio model, like, China is very much winning on the AI content side in part because of some of the decisions they've made. And we talked about that before. This is a interesting space in that, like, if China doesn't make people kind of like log in to do this stuff and they do have frontier models or at least one step behind the frontier model, that could shake up the distribution of who's winning. An AI kind of at large. So this is like a big. That was the second question that was.
- [10:09]
B
Yeah, that was to be the second question was that, you know, like, we talked years ago about an AI pause and it was like, well, you can't do that because if you pause, other nations are going to not only catch up, but they will leapfrog. And While there's probably 20 companies that don't mind because they still have access to this, the true frontier, like those foundational models, like, for everybody else, I'm starting to look at other countries a little closer and I'm starting to think, like, if they want to take a wax paper etching, if they want to do a stencil marking of my actual, that's fine, dude. If that will allow them to confirm my identity, they can put a little wax paper over it. You take a little lead pencil, you scrape, and it looks like a butterfly.
- [10:53]
A
That's a new fingerprint. That's why Anthropic's logo looks like that. Now we know. Now we know.
- [11:00]
B
Now you get it.
- [11:02]
A
Please spread to verify. No, no. I do think we're going to see over the next month to two months this kind of thing play out. I do assume there's a couple rumors out there that, like, next week, GPT 5.6 soul comes out. Who knows? At this point, we are very much in the hands of a government in America, at least. That seems kind of confused by a lot of this stuff. I feel like we're confused. But we will keep you up to date. As it happens. I am very excited to see how good 5.6 SOL is in comparison to Fable. Once we get both of them back, and we will try to benchmark them both against each other doing the dumb stuff that we do. But for now, we are in a holding pattern. Sonic 5 might get us to a better place, but I think that's kind of it on those for right now, which kind of blows, to be honest with you. That's a kind of a sucky update. But you know what? You want more sucky updates, you come right here, Kevin. You come right to this show. You click on subscribe, you click like you follow our Patreon. You give us money to do all the dumb stuff we do because we're still doing sucky updates every week. We're here every twice a week. Although not on the Friday show.
- [12:08]
B
It's fine, Gavin.
- [12:09]
A
We got a holiday.
- [12:10]
B
It's all good. We're jamming freedom dogs down our freedom gullets. It's all good. Are you there?
- [12:19]
A
Hello? Yeah, I'm here.
- [12:20]
B
Oh, okay. It's one episode this week, Gavin, but, you know, if that makes you feel some sort of way, drop a comment about it on the YouTube. We will read it and we will probably cry. We'll weep internally.
- [12:31]
A
But we're already weeping. We're weeping a lot.
- [12:34]
B
Never stop sobbing. That is our motto here. If you don't like leaving comments because it exhausts your little fingers. And I get that, Gavin. There might be hope on the horizon. What? If you could just think your toxicity into the corn pewter.
- [12:50]
A
That's what I'm doing right now. I'm thinking about what I feel about you. You right now living your best life in Spain. Can you read my mind?
- [12:58]
B
Oh, I thought you were thinking about my auth gate mechanism that I'm writing.
- [13:03]
A
No, I'm not thinking about that. That's right. So there's a big story out of Meta, which, by the way, not my favorite company to be doing this right now, but we can talk about that later. There's a new piece of science that came out of Meta called Brain to Qwerty, which again, thank you AI researchers for the best naming experiences I've ever. And Brain to Qwertity is a system that essentially is designed for people who have par paralyzer who have had a stroke to read their mind and allow them to type. And Kevin, the video. There's not a video. There's a kind of a research paper here, but there's a shot of them, like, in this chair with this, like, giant thing in their head. And it reminds me of the brain meme thing. Right? But it is a very funny thing. So essentially what this is doing is it's using AI to kind of like read the brain waves of real humans and then allow them to type. So this is a very cool piece of science, but when you start to extrapolate out further and further, and of course, as the Internet has done, this is a little bit of mind reading, right? Like the idea that we will be able to shoot each other thoughts without having to type or say anything that feels like we're getting closer to something that's very different than our normal way of living.
- [14:11]
B
I love it. This is going to be part of the new pre check. You're going to have to go in to your local government branch and do a calibration thing, right? So they know that they're reading your unique brain pattern successfully. And then when you walk through the, the rapid scan machine and you hold your hands up or whatever, they're going to just glean your thoughts, are going to know exactly what your true intentions are about your dear leader. And maybe you'll get on the flight or maybe you'll be incinerated. Our future is going to be beautiful and perfect. Why are people up in arms about this?
- [14:41]
A
Well, you know why? It's because it's meta. And Meta has a bad history of advertising, tracking and all sorts of other things. And of course they've also been trying to train their AI models, people who work there. And people have not been very thrilled about that. But Kevin, I think overall this is an interesting like thing where like the science of this is so exciting but then the practical applications are very scary. So one thing to keep in a, to keep tracking with your brain or without? I'd like to try to without for right now. I just got to let it float off in a bit. Another quick story. Nano Banana 2 Lite has been released. Google is still pumping out the models in the cheaper kind of faster world of AI. There's some really interesting stuff with this model. It's not as good as nanobanana 2 or of course Nano Banana Pro, but it is fast and is cheap. And you always say that triangle right of good fast or cheap? This is fast and cheap and it's not going to be like, you know, frontier model images, but if you have an, an idea or if you have a project that needs just kind of quick images generated fast, like it is very good. There's a very cool video somebody released about like how quickly it could generate images. So I think that was a very, very cool use case of Nano Banana. And you know, they're still pushing forward. We assume there will be a new Nano Banana maybe when Gemini 3.5 Pro comes out, but we will see where we go from here.
- [15:59]
B
Now, Gavin, you have the AI for Humans Creative MCP Festival. What?
- [16:05]
A
That's right.
- [16:06]
B
Is that a typo? Because, I mean, I know ICP has a gathering, which I think is phenomenal.
- [16:13]
A
This is the AI for Humans Creative MCP Festival. That's right, everybody. That's right. Welcome to the party. This is not the icp. This is MCP Master Context Protocol. Right. Is that what MCP stands for or. No, something else.
- [16:26]
B
A model Context Protocol.
- [16:28]
A
Master Context Protocol is the thing from Tron, So too close to the Tron name. But anyway, what Kevin and I wanted to do this week was show you all a couple really interesting workflows that we're using with mcp. So we've talked about this last week and a couple times in the recent past. MCP is a connective tissue that allows you to connect Claude or OpenAI directly to the back end of a particular soft piece of software. And Kevin wants to talk a little bit about this thing that we both saw blow up was connecting the MCP to Blender, which is a very cool AI 3D modeling software. Tell us a little bit about this and kind of what you did.
- [17:05]
B
Yeah. So Blender, Dead simple. Well, I shouldn't say dead simple, but very popular. It's super complicated. It's actually very, very complicated. But method is dead simple. Blender allows you to arrange cameras and 3D objects in a scene. You can make really, really detailed, beautiful stuff with it, but you can also rough out very primitive stuff fairly quickly.
- [17:26]
A
Right.
- [17:28]
B
The MCP of this all allows you to connect Blender to cloud code or Codex or Gemini, whatever tool set you are using and experimenting with. And you can just ask your agent to assemble a scene for you. Give me a blue block with a camera that starts very far away and dramatically tilts down and goes towards blue block and then starts to shake violently. You then export that viewport, right? Just that window of the camera movement with the primitive geometry. And you put it into something like Seed Dance. And you say the blue block is actually a breakdancing clown at the gathering of the MCP Festival. And the video model then replaces that rough geometry with really high def visuals. And what people are doing is astonishing. They're getting really, really gran creative control over the way the cameras move, the way characters interact with each other, or complex geometry within a scene. And it's all because of these little blocky things. So I started working on a video of my own. Gavin and I got about as far as generating the blocks in the viewport. I haven't quite figured out exactly what it's going to be, but I'll release the files to our patrons and if they want to take it and do something with it.
- [18:37]
A
Yeah, why don't I. You haven't sent that to me. So while you're sending that to me, let me give you a couple updates on what this is. So, first of all, this guy named Reese Reed Hannaford was the first person that I saw do this. I think it's something that happens. But we will link his kind of workflow. He generates the start frame in mid journey. So there's a mid journey start to this. And then he blocks out the action in Blender. And what's cool about this with Blender is it's really about camera control. So, like, you can control where you want the camera to go. So in addition to the animation of the blocks moving across, like, one of the examples he has is, like this very cool shot of a train where next to the train, there's a guy on a horse walking, you know, running along the train. You then are able to basically take that Blender output, which is, you know, essentially like a childhood stick moving, and then use that to drive the animation of the sea dance or whatever prompt. And what it does is it just gives you much greater control over the actual output. It's very cool. So, Kev, let me look at this thing that you are creating. Oh, oh, oh, oh. Yeah. I'm thinking, like, I don't know about this guy.
- [19:39]
B
I'm thinking like, Tetris. Tetris blocks are like us. Like, what are they doing on the off time? Maybe they're doing laundry. I don't know. Or maybe it could be like a horse drinking from a trough. And I'm using some placeholder artwork. There's a lot that we could do with this, Gavin.
- [19:52]
A
Sure, sure. Kevin's got.
- [19:53]
B
Look, it's a young workflow.
- [19:55]
A
Go to our Patreon if you want. It's a good example of what you could actually get there. It's the first time we'll have updated our Patreon with new content for a very long time. So that is a very cool thing at large.
- [20:05]
B
Yeah. This is a whole MCP festival, so we can't just have one vendor.
- [20:08]
A
Right.
- [20:09]
B
This isn't like a celebration of America. We have to actually have a packed house. ComfyUI has an NCP integration now. If you've ever messed with ComfyUI. Even if you have it, you know it as spaghetti wires connecting bricks to other bricks. You can do the most insane, cutting edge stuff, but it is very complicated. A lot of the advanced workflows require. I mean, there's so many boxes and little wires to connect and whatever. But now they have an MCP server. So you can ask your agent to do a lot of the research and do a lot of the heavy lifting for you and building out workflows that give you effects that you want out of the software, essentially.
- [20:44]
A
That's right. And the biggest thing with Comfy is that you can really get granular on control where you can't do that sometimes with other things. So even if you're plugging in a model that you're paying for, you can do a lot more, you can export a lot more, and especially locally you can do a lot. But I will say this is a big deal because Comfy is a confusing sort of scenario. It's a very cool thing to go try. It is out now, they've just fully launched it. And then, Kevin, there's one other MCP that I do think is worth talking about, which is the X mcp. And what we've talked about for a while is that X has this amazing data set behind it, formerly Twitter, where all these people are talking in real time about all this stuff and they have released an MCP to access it. But do you know what the downside of this is, Kevin? It's not like it's unlike other MCPs.
- [21:25]
B
Oh, how so, buddy?
- [21:27]
A
Well, this is a very expensive MCP to use. This is not free. In order to use the XAPI which drives this mcp, you end up having to pay quite a bit of money. So all the fun stuff that you think you would want to do with X and X accounts or like learning about what people are posting, all that stuff is not free. And I do think this is a world where we're going to enter the space of like, I understand these companies wanting to control their data, especially companies like X that are trying to train models on what we all write on there. But like there's a level of the cool stuff that could be built by the masses that is just not going to get built when you have an expensive MCP like that. I think it's like, like it's like 45, $5,000 to kind of have the entry level access to their API, which is a lot of money. So like this is just gonna stop
- [22:12]
B
looking at people from using it, looking at like, just post read, right, where it's 0.005 cents per resource. Like an individual tweet at that point, or post on X could be a resource. And if you're trying to scrape from that massive fire hose, where you and I both know a large portion of it is bots and AI generated slop and nonsense. And so it's just now they're going to be further disincentivized to actually clean up their environment because all of that activity is going to cost somebody to read and sift through or write against or mute or block or whatever. All of those actions are now going to cost fractions of a cent. But when you start doing them across millions of tweets globally, I mean, look, I. I really want the platform to clean up and do better. I just don't know that this is going to help it.
- [23:01]
A
Yeah, I mean, I think we'll see. I think this is going to be an interesting pathway to what there is. I do think there's a level of like, at least being able to read these tweets that we got to figure out some way to do. But, yeah, I mean, I agree it's a tricky thing. Before we go, I do want to shout out. I've been working on a really interesting micro drama workflow using C dance 2 that I posted about in our newsletter this week. Go follow that if you can. We'll drop a link in the show notes, but I want to spend more time working on this. And Kevin, I'm going to spend some time over this week that we're off and really try to come out of this week with a, you know, a 10 episode, two minutes per episode microdrama and just try to make it, because I think it's going to be really interesting to kind of have this experience. So I'll tell everybody about that more when we get back. And for now, I have a wonderful.
- [23:43]
B
I have some characters, by the way, Gavin, that need some backstory. There's a purple block and a red block and then there's a.
- [23:49]
A
Not my micro drama. You can create your own micro drama.
- [23:52]
B
I just want to be clear. It doesn't have. But I'm open. Gavin. It doesn't have to be a washing machine, right? It could be. Be a dishwasher, could be.
- [23:58]
A
Oh, it's like a Pixar movie. It's like. Is that the kind of thing. It's like, you know, all the appliances in my.
- [24:03]
B
You can get stuck in all sorts of stuff. Oh, no, no, no. Bye, everyone.
- [24:08]
A
All right. We'll see you all. See you all a week from now. Hopefully by the time we come back, we'll have access to these amazing frontier models. We'll see. Y' all have a nice Fourth of July in America.
- [24:16]
B
Bye.
- [24:17]
A
Bye, y'.
- [24:17]
B
All. Bye.