I have heard people talk about the overall number of works online (let’s say about 20 billion) and that rather large corpus of words being used for training large language models. My personal writing collection is over five million words and was used to train a custom GPT-2 model to write and sounds an awful lot like my personal writing style. Currently my GPT-4o model that I interact with regularly does not use that foundation, but is capable of mimicking my writing style on command. I just ask it to write in the style of “Nels Lindahl” and it does a pretty good job of producing prose that is very similar to my standard writing voice. Over the years my speaking voice and writing voice are not all that different. I do think my written style is more verbose and for the most part contains a higher quality of communication. When you are talking to somebody the expectation is that you will not speak in 1,000 word blocks of thought. You have to be a lot more word-economic during the conversation. 

To be fair about the previous consideration almost all conversation is subject to rules that enforce some type of compression compared to the written word. Conversations and dialogue that happen during a meeting have such a short timeline where getting to agreement is either directed by one part or the other saying this is the way or during a truncated debate where because the timer is running agreement has to be reached within the artificial confines of the meeting or more meetings will ensure. That is one of the strangest groupthink drivers of complex workplace interactions. It’s like the deciding elements of things are spread out and people talk about it and sometimes decisions are made, but how are those decisions then communicated back out in a way that is actually heard and understood to be implemented. For simple things it probably is easy to have an affirmative or negative outcome shared with the group. For really complex things that decision is translated and navigated a lot more than the decider presumed. 

Sometimes a memo or extensive direction could be sent out to really communicate a complete decision. That is very rare in my experience. It’s interesting that we spend very little time thinking about the science of how decisions are made and what that means for the ongoing operation of a business. More time should be spent into how the trajectory of decisions and the outcomes of those decisions are spread throughout modern organizations. When a lot of the street level decision making in an organization is communicated by chat based interfaces like Slack, Microsoft Teams, or Google Chat. I actually think one of the reasons that those systems are favored by smaller focused teams is that it is easier to have channels where decisions and directives are communicated. Tracking and understanding methods of communication in the modern office is an interesting place to force an intersection with considering the practical applications of decision making and really dig into what is happening.  

We are probably facing a future where agents deployed in workplace environments are designed to be able to evaluate communication and extend the directions of leadership and executives in terms of a sort of monitoring service across the daily work of the organization. You would have the executive monitoring agent sort of provide soft commentary that maybe this direction or that previous outcome should be considered based on what is currently being proposed. This would create a lot more ongoing continuity of decision making, but it might also expose situations where the executive teams zigs and zags a lot into conflicting directions. Over time based on new information it is possible to move from one position of argument to a new and potentially opposite one. Generally speaking zigging and zagging on a more routine basis than what could be considered over time could be highly confusing as it takes awhile for a decision to be communicated anyway.

Dr. Nels Lindahl