Large language models lie because they're trained on messy data and rewarded for sounding confident. New Constructs built a smaller AI focused only on financial fundamentals, using a curated dataset they claim is 100% accurate. Google partnered with them to create FinSights, an AI investing agent that trades accuracy for usefulness.
Chapters
And I challenge everybody, you know, that's able to listen to this, go into Gemini or to Claude or to ChatGPT and ask it, why does AI lie so much? It'll give you 5 to 7 reasons. And those reasons are always, well, my training data is not complete or accurate. I'm trained to give a response and show confidence, not admit that I am wrong. Any answer is greater than no answer. Anyway, and it shouldn't be confident to extrapolate all these things. What no one talks about, about AI is that the data is not any good. Nobody knows how to teach it. So Google actually paid New Constructs to have access to our data, to access to my team, to build its own AI agent for investing. It's called FinSights. They featured it last week out in Vegas. And they brought us on to give them what I call a small language model, our small language model. small language model is very different from the large language model in that its ambitions are much more practical. Our ambition at New Constructs is to understand fundamental data, a silo, not the whole world. We're not trying to boil the ocean. We're focusing on the silo of fundamental financial data. And that's something that we've been able to use technology to make sure our data is as accurate as possible, which is 100%. If you ask Harvard Business School, published a paper proving that. And that, that data is also organized into an ontology that generates alpha, which Harvard Business School has proven, which Sub-Zero has proven, which our publicly traded indices, Bloomberg, have proven. So that's why Google said, listen, if we can combine accurate dataset with verifiable ability to generate alpha into an AI agent, now we have something really powerful. And that's why they built it.
