Reasoning Chains for Research

Many of us are using AI to perform research, to do web searches, to use our own data, to merge a number of different sources, both on the web and something that maybe we have locally, in order to synthesize and make decisions. Now for consequential work, for the work you do that maybe impacts your business or bottom line, it's pretty important that you, feel confident that the facts that you're pushing, the information that you are going to include in any reports that end up in the hands of executives, stakeholders, or customers is trustworthy, is true, is factual as much as possible. That's

what makes reasoning chains so powerful in count. So in this case, I've asked the count agent to research this year's biggest consumer tech software and hardware announcements. It's used one of Anthropics models in order to generate this report right here, which includes a number of...

If I go over to lineage and sources, you can see all of the different information that it pulled from the web in order to draw the conclusions that it drew. I'm going to click on the hero finding here, which includes many of the assumptions and sources cited. But you can see that mostly if I click on an article, it'll show you the text. If I click on a quote from the article, it'll show you where you got it.

So you really can use reasoning chains to trace information back, but reasoning chains can also reveal information that maybe Claude or Chad GPT wouldn't, and that's when there's contrary evidence, there's problems with the reasoning. In this case, in the report, it cites the steam frame as having one price. In this case, we have conflicting pricing. It's basically a signal to us to go double check that.

So in this case, I will ask the agent to update this.

Also within the lineage tab here, you can see assumptions at the highest level and the primary issues within reasoning presentation, basically the biggest issues that need to be addressed before you take any research and put it out to the world. Anytime you ask the agent for more information, you supply new information, you update anything up and downstream, it will rereview itself, meaning that you're always left with something fresh and more accurate than the last report. So there you have it. Research and count using reasoning chains in order to generate reports and findings that are accurate, factual, and hopefully, hallucination free.