00:00 Speaker A
My experience with Open AI Claude is that if you have the latest model and you tell it to work and put it in thinking mode, you don’t have to worry about hallucinations, but is that just kind of avoiding the problem? How are you attacking this?
00:21 Speaker B
Yeah, III I mean, I’m glad to hear you’re having a good experience. I think this kind of model can be set up to minimize hallucinations, but it’s still going to make all sorts of mistakes. And I think one of the interesting things about these frontier models is that if you ask a question and get an answer, it might be right all along. But if you ask again, perhaps just in a slightly different way, you might get a completely different answer. And this is really fundamentally indicative of what some people call spike intelligence, or something like a lack of determinism in these models. They just can’t reliably give the right answer over and over again. And that’s exactly the problem that scale awareness is trying to solve.
01:21 Speaker A
So how do you approach this? Is this a structural problem? What makes you different from your competitors? And I just wanted to ask you, is your competition like Open AI or, um, Anthropic?
01:38 Speaker B
So I think what sets us apart is that we took a radical approach from the beginning. We are Frontier Model Research Institute. We have a large team of PhD researchers with extensive experience in this field, as well as extensive experience in applying this technology on a large industrial scale. Well, I worked at Microsoft, somewhere like that. So we knew that the very large language modeling systems out there had fundamental limitations in their ability to perform tasks reliably and repeatedly. So we knew we needed to fundamentally change the way these systems were built. So we took a radical approach and completely changed how the model is trained, its internal techniques, algorithms, architecture, the so-called loss function, and the data that is input to it. The systems we build are actually a different animal. Now, I would say that one of the things that makes us different and gives us a little bit of an edge is that we’re focused on these enterprise applications. Our model is not designed to power consumer applications. For example, like ChaBT, you should be able to answer questions about everything. And that focus gives you some technical angles that you can take to actually solve problems in a corporate environment.
