Langbase’s Post

Transcript

What is going on? I know what is going on just 10 minutes ago opening I launched their new series of models code name 01. These are super advanced models launched as beta models beta models like and their reasoning models. I just infect went ahead and used all one mini model to create a reasoning agent. This reasoning agent is live online base. In fact, if you have Tier 1 tier 5 access on open AI, you can build a reasoning agent with one on land based today right so I went ahead and picked. And many and in this reasoning agent, what I'm building is a React component. I gave it a React component a pretty simple one. And what I'm asking is you're gonna change this for all the non fiction books in seeing the text to read, right And then another thing I want is for it to only return code after that for formatting. I'm asking it to return code that has four spaces as steps and it should not allow the code, the line of code to exceed more than 80 columns. So it should be pretty smart right about all of this. So I think what it's gonna do is it's. Do a chain of thought over all of this and generate the perfect high quality code that follows my instructions to the letter, right? So, and it takes a bunch of time. These models are really slow. I think it's a recursive loop in there, but I'm gonna just run it and go back and show you. If you go to ourmodelpageslanguage.com models, you can see all the models here. Let's select open AI and let's go with 1 preview. So let's see what this model is all about. SO1CS models are new alms and they're trained to reinforcement learning. To solve complex reasoning right this is you know solve hard problems through complex reasoning. They think before they answer and can produce long internal chain of thought before responding to the user. The excel at scientific reasoning ranking 89th percentile on this coding programming question code forces and it's pretty good like you can see all they can potentially sometimes exceed human PhD level accuracy. I don't know but it sounds really good right anyway, they were trained on making. Where you know the disallowed content, training data, host edition buys. All of this is covered, right? And this is the scorecard they have around all of this. I quickly took a look around, read all of the content in our head of AI. And you know, the team built up a recommendation for all of you around where this new series of oven models you can use, I think for workflow automation, especially related to code or scientific stuff, just like they said for science and math and chemistry. Even biology, I haven't really tested this far. But for coding questions, we did test. In fact, let's go and see what's happening here. But it has already come up with a response. So this is how GPT 4 would respond. It would not care about how lengthy this particular line is. But here is 1 mini. And if you take a look at this, this is pretty amazing, right? So if I were to check it out 1234 yes, the spacing is correct. And this is a lengthy line. As you can see, it divided it into multiple set of lines. So it would not go beyond 80 columns in code. And this is the older models that wouldn't just care about things like this at all, right. So this is how. Would these alms are these new models are at following your instructions to the letter, right? So if you're building some complicated chain of thought workflows, I think you can try out these models. This could really, really high compared to others on the benchmarks. Like I was looking at this particular coding benchmark where compared to GPT 4 or the performance of 1 preview is 62% on completing code on code completion and one which they are probably going to launch soon since this is open preview is even better 89%. I am really excited. I think it's it's it's the fall, right? And you know, everybody's gonna all these model creating companies, all these providers are going to launch their best of the best state-of-the-art models today or within a month or so. And it's just an exciting time. Go ahead. Check out their transparency, check out why you should be using these models. I don't think that one is here to replace DVD for I think it will augment a multi step workflow where you will reason through an oven model and then use that. Response to do a bunch of different things and compose a complicated pipeline. You can literally go ahead, pick oven preview or one mini and you can build a composable ancient complete pipeline that you can deploy online based. And literally like for example, on our home page you will see us explain the concept of composability. Let's say we get an e-mail, a set of different agents will run and all of these are live pipes with APIs, serverless, and I think we can use super looms. And just, you know, doing sentiment analysis or summary for most of this case. And for that, the scene maker, this one we used to use DPD for, for this this kind of agent. But now I think this team of thought the scene making where we have a bunch of different instructions. This scene maker is going to be the best candidate for using these open models where it will use whatever products we have for this particular LLM and it will make sure that this is a frustrated customer and they need to. You just want it soon and then they will pick probably something like for many visits cheaper to write an e-mail back and you can literally open up this playground right here and play around with Owen online base today right? Pretty excited. Let me know what you think. Let me know what you are building, what you are going to ship with open AIS over take care Ciao bye.